JP6160438B2 - Machine translation apparatus, machine translation program, and machine translation method - Google Patents

Description

  The present invention relates to a machine translation device, a machine translation program, and a machine translation method that machine translate a sentence in a first language into a sentence in a second language.

  2. Description of the Related Art Conventionally, machine translation devices and machine translation software that translate a sentence in a first language to be translated into a sentence in a second language different from the translation target are known. These machine translation devices and machine translation software translate a sentence in the first language into a sentence in the second language using a bilingual dictionary of the first language and the second language.

  However, if a sufficient number of words are not registered in the bilingual dictionary, the frequency of occurrence of unknown words (unregistered words) increases in the first language analysis step, and high translation accuracy cannot be obtained as a whole. Therefore, in order to improve the translation quality of machine translation, it is necessary to enrich the bilingual dictionary. Many techniques for extracting unregistered words from documents in the field to which machine translation is applied have been disclosed for the purpose of eliminating unregistered words. For example, a technique is disclosed in which hiragana unknown words are extracted in units of words, focusing on the fact that a single hiragana character is extracted as a word when an unregistered word is included in the hiragana part of Japanese text (for example, patents). Reference 1).

  However, even if many unknown words can be extracted by the unknown word extraction technique, the accuracy of machine translation alone cannot be improved. For example, in the case of machine translation from Japanese to English, a translation (English) must be registered by a human for each unknown word list (Japanese) extracted by the unknown word extraction technique. The task of registering this translation is not only fluent in both Japanese and English, but often requires specialized knowledge in the field, which is expensive in terms of manpower and man-hours. End up. In addition, the maintenance of a bilingual dictionary is required for each language pair, and the work for maintaining a bilingual bilingual dictionary requires much more man-hours.

  In addition, as a technology to reduce the quality of translations caused by unknown words by reducing the manual translation dictionary, the first and second language bilingual dictionaries have been expanded using the first and third language bilingual dictionaries. The technique to do is disclosed (for example, refer patent document 2).

  FIG. 1 is a diagram for explaining a technique for expanding a bilingual dictionary for a first language and a second language using a bilingual dictionary for a first language and a third language.

  In FIG. 1, there are only Japanese “decline” -Chinese “fall” in the bilingual dictionary (Japanese-Chinese dictionary) of the first language (Japanese) and the second language (Chinese). The bilingual dictionary (Japanese-English dictionary) of the first language and the third language (English) includes Japanese “drop” —English “drop”, Japanese “drop” — “drop”, Japanese “low drop” — "Drop", Japanese "drop"-"drop" exists. In such a case, search for “drop” using a Japanese-English dictionary from “drop”, and then search for “drop”, “low drop”, and “drop” from “drop”. Ask for. By registering the retrieved synonyms “fall”, “fall”, and “drop” in association with Chinese “fall”, “descent” − “fall” that was not included in the daytime dictionary. Can be supplemented. By expanding the bilingual dictionary in this way, it is possible to translate “descent”, which was impossible to translate from the first language (Japanese) into the second language (Chinese), as “falling down” Become.

JP-A-1-220067 JP 2007-26420 A

  However, even if you try to expand the bilingual dictionary for the first and second languages using the bilingual dictionary for the first and third languages, there is only one bilingual translation in the bilingual dictionary for the first and third languages. In some cases or in the case where none exists, the bilingual dictionary of the first language and the second language cannot be expanded, and high translation accuracy cannot be obtained as a whole.

  In one aspect, the present invention can obtain high translation accuracy as a whole even when there is only one or no parallel translation in the bilingual dictionary of the first language and the third language. An object of the present invention is to provide a machine translation apparatus, a machine translation program, and a machine translation method.

The present invention employs the following configuration in order to solve the above problems.
In one proposal, the machine translation device registers an input unit for inputting a sentence in a first language to be translated and at least one second language expression in association with at least one expression in the first language. Morphological analysis of the first language second language parallel translation dictionary and the first language text input by the input means, and referring to the first language second language parallel translation dictionary, the first language Is a first language second language translation means for translating into a second language sentence for different translation purposes, and the first language second language based on the translation result and the morphological analysis result by the first language second language translation means. An unknown word recognizing means for recognizing an unknown word in which the expression of the second language is not registered in the bilingual dictionary; and at least one expression of the first language is different from the first language and the second language One third language expression Correspondingly registered first language third language bilingual dictionary and morphological analysis of the first language sentence input by the input means, referring to the first language third language bilingual dictionary, A first language third language translation means for translating into a third language sentence, and an unknown word recognized by the unknown word recognition means among the third language sentences translated by the first language third language translation means An unknown word third language cutout means for cutting out a third language corresponding to a word; a third language morpheme analysis means for dividing the third language cut out by the unknown word third language cutout means into a plurality of morphemes; A third language first language dictionary lookup means for searching for a word in the first language corresponding to the divided morpheme with reference to a first language third language parallel translation dictionary; and the third language first language dictionary lookup. Using means An alternative notation generating means for generating an alternative notation of the unknown word recognized by the unknown word recognizing means by combining a plurality of words of the first language corresponding to the plurality of divided morphemes, and input by the input means Of the inputted sentences in the first language, the unknown words recognized by the unknown word recognition unit are replaced with the alternative notation generated by the alternative notation generation unit, whereby the inputted first language sentence The first language second language translation means translates the first language sentence corrected by the input sentence correction means into a second language sentence. .

  According to the present invention, by creating an alternative notation for an unknown word that cannot be translated because it is not registered in the bilingual dictionary, the machine translation quality is greatly improved.

It is a figure for demonstrating the technique which expands the bilingual dictionary of a 1st language and a 2nd language using the bilingual dictionary of a 1st language and a 3rd language. It is a figure for demonstrating the outline | summary of a machine translation process. It is a figure which shows the structure of the machine translation apparatus 1 of 1st Embodiment. It is a figure which shows the example of the 1st language 2nd language parallel translation dictionary. It is a figure which shows the example of the 1st language 3rd language parallel translation dictionary. It is a flowchart which shows the flow of an unknown word 3rd language extraction process. It is a figure for demonstrating the operation example of a machine translation process (the 1). 6 is a diagram illustrating an example of an unknown word list 800. FIG. It is a figure which shows the example of the 1st language morpheme list. It is a flowchart which shows the flow of an alternative description production | generation process. It is a flowchart which shows the flow of an input sentence correction process. It is a figure for demonstrating the operation example of a machine translation process (the 2). It is a figure which shows the example of the 1st language morpheme list | wrist 1300. FIG. It is a figure which shows the structure of the machine translation apparatus 2 of 2nd Embodiment. It is a figure which shows the structure of the machine translation apparatus 3 of 3rd Embodiment. It is a figure which shows the structure of 4th Embodiment. It is a figure which shows the structure of 5th Embodiment. It is a figure which shows the structure of the machine translation apparatus 4 of 6th Embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
The first embodiment will be described with reference to FIGS.

First, the outline of the machine translation process in the present embodiment will be described with reference to FIG.
FIG. 2 is a diagram for explaining the outline of the machine translation process.

  The example shown in FIG. 2 shows a case where the word “object oriented” is included in a first language to be translated, for example, a Japanese sentence. And when translating into a second language for translation purposes different from the first language, for example, a Chinese sentence, Chinese corresponding to “Object Oriented” in Japanese may not be registered in the bilingual dictionary. is there.

  In such a case, the machine translation process in the present embodiment executes translation into a third language, for example, English, which is different from the first language and the second language. Then, English “object oriented” corresponding to Japanese “object oriented” is searched. Then, the English “object oriented” is divided into two words “object” and “oriented”.

  Next, using the English-Japanese bilingual dictionary, the Japanese corresponding to the divided “object” and “oriented” is searched. For example, as for Japanese corresponding to English “object”, three of “object”, “object”, and “purpose” are searched. Also, Japanese “oriented” is searched for English “oriented”.

が検索され、日本語「指向の」に対応する中国語「面向的」が検索される。なお、日本語「物体」「目的」に対応する中国語は、日中の対訳辞書に登録されておらず、検索されない。 Next, using the bilingual bilingual dictionary, the Chinese corresponding to the searched Japanese “object”, “object”, “purpose” and “oriented” is searched. As a result, the Chinese corresponding to the Japanese "subject"

Is searched, and Chinese “face-to-face” corresponding to Japanese “oriented” is searched. Note that Chinese corresponding to Japanese “object” and “purpose” is not registered in the bilingual dictionary during the day and is not searched.

と、日本語「指向の」に対応する中国語「面向的」を組み合わせて、日本語「オブジェクトオリエンテッド」に対応する中国語として、

が出力される。 And Chinese corresponding to Japanese "subject"

And Chinese corresponding to Japanese “Object Oriented”, combined with Chinese “face oriented” corresponding to Japanese “Oriented”,

Is output.

  FIG. 3 is a diagram illustrating a configuration of the machine translation apparatus 1 according to the first embodiment. FIG. 4 is a diagram illustrating an example of the first language / second language parallel translation dictionary 13. FIG. 5 is a diagram illustrating an example of the first language / third language parallel translation dictionary 16.

  In FIG. 3, the machine translation device 1 includes an input unit 11, a first language / second language translation unit 12, a first language / second language parallel translation dictionary 13, an unknown word recognition unit 14, a first language / third language translation unit 15, A first language third language parallel translation dictionary 16, a third language morphological analysis unit 17, an unknown word third language segmentation unit 18, a third language first language dictionary lookup unit 19, an alternative notation generation unit 20, an input sentence correction unit 21, And an output unit 22.

  The input unit 11 inputs a sentence in a first language to be translated (for example, Japanese. Hereinafter, sometimes simply referred to as “day”).

  The first language / second language translation unit 12 performs morphological analysis on the sentence of the first language (day) input by the input unit 11, and refers to the first language / second language parallel translation dictionary 13 as shown in FIG. Thus, it is translated into a sentence in a second language (for example, Chinese. Hereinafter, it may be simply referred to as “middle”) for translation purposes different from the first language.

  In the first language / second language parallel translation dictionary 13, at least one expression of the second language (medium) is registered in association with at least one expression of the first language (day).

  Based on the translation result executed by the first language / second language translation unit 12 and the result of the morphological analysis, the unknown word recognition unit 14 stores the representation of the second language (medium) in the first language / second language parallel translation dictionary 13. Recognize unregistered first language (Japanese) expressions as "unknown words".

  The first language / third language translation unit 15 performs morphological analysis on the sentence of the first language (day) input by the input unit 11, and refers to the first language / third language parallel translation dictionary 16 as shown in FIG. Thus, the sentence is translated into a sentence in a third language (for example, English. Hereinafter, it may be simply referred to as “English”) different from the first language (Japanese) and the second language (middle).

  The first-language third-language bilingual dictionary 16 has at least one third language (English) different from the first language (day) and the second language (medium) for at least one expression of the first language (day). ) Is registered in association with each other.

The third language morpheme analyzer 17 divides the third language (English) sentence into a plurality of morphemes.
The unknown language third language cutout unit 18 corresponds to an unknown word recognized by the unknown word recognition unit 14 among the sentences of the third language (English) translated by the first language third language translation unit 15. Cut out the language (English). Then, the unknown language third language extraction unit 18 divides the extracted third language (English) into a plurality of morphemes using the third language morpheme analysis unit 17.

  The third language first language dictionary lookup unit 19 refers to the first language third language parallel translation dictionary 16 and searches for words in the first language (day) corresponding to the divided third language (English) morphemes. To do.

  The alternative notation generation unit 20 uses the third language first language dictionary lookup unit 19 to each of a plurality of morphemes divided by the unknown language third language extraction unit 18, that is, a plurality of words in the third language (English). Search for words in the corresponding first language (day). And the alternative notation production | generation part 20 produces | generates the alternative notation of the unknown word recognized by the unknown word recognition part 14 by combining the searched several word.

  The input sentence correction unit 21 replaces the unknown word recognized by the unknown word recognition unit 14 with the alternative notation generated by the alternative notation generation unit 20 in the sentence of the first language (day) input by the input unit 11. Thus, the input sentence of the first language (day) is corrected.

  The output unit 22 outputs the sentence in the first language (day) corrected by the input sentence correction unit 21. The corrected first language (Japanese) sentence that has been output can be translated with higher accuracy by the first language / second language translation unit 12.

  In the above description, the first language to be translated is Japanese, the second language to be translated is Chinese, and the third language is English. However, the first language, the second language, and the third language are described. There are no restrictions on the combinations.

  Unknown words are classified into proper nouns, technical terms, common nouns, and the like. Technical documents such as manuals for machines, tools, application software, etc. have many unknown terms. In general sentences such as letters, there are many unknown words such as proper nouns and common nouns. Also, in Japanese, there is a custom of making a new foreign word katakana, and if it is not registered in the dictionary, that katakana is an unknown word. In a broad sense, compound words are a kind of unknown word. For example, “machine translation” is one word, but if there are the words “machine” and “translation”, a meaningful translation can be made.

  The machine translation process executed by the machine translation apparatus 1 will be described in detail with reference to FIGS. The machine translation process includes an “unknown word third language extraction process”, “alternative notation generation process”, and “input sentence correction process” which will be described later.

  FIG. 6 is a flowchart showing the flow of unknown word third language extraction processing. FIG. 7 is a diagram for explaining an operation example of machine translation processing (part 1). FIG. 8 is a diagram illustrating an example of the unknown word list 800. FIG. 9 is a diagram illustrating an example of the first language morpheme list 900.

となる。また、翻訳の過程で実行される形態素解析の結果として、原文形態素７０３「/彼/は/オブジェクトオリエンテッド/ランゲージ/を/勉強/し/て/いる/。/」を取得する。 In step S601 of FIG. 6, the sentence in the first language (day) input by the input unit 11, for example, the original sentence 701 in FIG. 7, “He is studying object-oriented language” The bilingual translation unit 12 translates. The translation result is a Japanese-Chinese translation 702

It becomes. Also, as a result of the morphological analysis executed in the translation process, the original text morpheme 703 “/ hi / ha / object-oriented / language /// study / do / de ///” is acquired.

と、取得した原文形態素７０３「/彼/は/オブジェクトオリエンテッド/ランゲージ/を/勉強/し/て/いる/。/」に基づいて、第一言語第二言語対訳辞書１３に登録されていない第一言語（日）の表現、未知語７０４「オブジェクトオリエンテッド」を「未知語」として認識する。そして、認識した未知語を、図８に示すような未知語リスト８００に作成する。 In step S602, the unknown word recognition unit 14 translates the Japanese-Chinese translation 702.

And is not registered in the first language bilingual bilingual dictionary 13 based on the acquired original text morpheme 703 “/ he / ha / object oriented / language /// study / do / te / is /./”. The expression of the first language (day), the unknown word 704 “object oriented” is recognized as “unknown word”. Then, the recognized unknown word is created in an unknown word list 800 as shown in FIG.

  In step S603, the first language third language translation unit 15 translates a sentence in the same first language (day) as the sentence in which the first language is translated into the second language in step S601, and obtains a morphological analysis result. . In other words, the first language third language translation unit 15 translates the sentence 701 “he is studying object-oriented language”, which is the first language (day) sentence input by the input unit 11. The translation result is a Japanese-English translation 705 “He is studying object oriented language.” Also, as a result of the morphological analysis executed in the translation process, the original text morpheme 703 “/ hi / ha / object-oriented / language /// study / do / de ///” is acquired.

  In step S604, as a result of the morphological analysis performed by the first language / second language translation unit 12 in step S601, the original morpheme 703 “/ he / has / object-oriented / language // study / do / do / is /. The first language morpheme list 900 as shown in FIG. 9 is created based on “/”.

  Then, in order to make the unknown words registered in the unknown word list 800 the target of processing in step S606 and subsequent steps, in step S605, it is determined whether or not the end of the unknown word list 800 created in step S602.

  If it is determined that the end of the unknown word list 800 is reached (step S605: Yes), the unknown word third language extraction process is terminated. On the other hand, when it is determined that it is not the end of the unknown word list 800 (step S605: No), step S606 and subsequent steps are repeated in the order registered in the unknown word list 800.

  In step S606, it is determined whether or not the end of the first language morpheme list 900 created in step S604. This determination step is for the first language morphemes registered in the first language morpheme list 900 to be sequentially processed in step S607 and subsequent steps.

  When it is determined that the end of the first language morpheme list 900 is reached (step S606: Yes), the process returns to step S605, and the next unknown word registered in the unknown word list 800 is set as an object of processing from step S606. On the other hand, when it is determined that it is not the end of the first language morpheme list 900 (step S606: No), step S607 and subsequent steps are repeated in the order registered in the first language morpheme list 900.

  In step S607, it is determined whether or not the unknown word in the unknown word list 800 and the first language morpheme in the first language morpheme list 900 are the same.

When it is determined that they are not identical (step S607: No), the process proceeds to step S608.
In step S608, it is determined whether or not the unknown word is included in the first language morpheme. If it is determined that the unknown word is not included (step S608: No), the next first in the first language morpheme list 900 is determined. The process returns to step S606 with the language morpheme as a processing target. That is, they are not the same (step S607: No) and are not included (step S608 :) when the unknown word and the first language morpheme are completely different. On the other hand, if it is determined in step S608 that it is included (step S608: Yes), the process proceeds to step S610 described later with reference to FIG.

  On the other hand, if it is determined in step S607 that they are the same (step S607: Yes), in step S609, the third language notation corresponding to the first language morpheme is set as the third language notation corresponding to the unknown word. Store in language notation list. Then, the next first language morpheme registered in the first language morpheme list 900 is set as an object of the processing after step S605.

  Here, the flow of the above steps S605 to S609 will be described more specifically.

  In step S605, it is determined whether or not the end of the unknown word list 800. In the unknown word list 800, as shown in FIG. 8, one unknown word 704 "object oriented" is stored. In the first step S605, the first unknown word 704 “object-oriented” in the unknown word list 800 is a processing target. It is determined that it is not the end of the unknown word list 800. Therefore, it progresses to step S606 (step S605: No).

  In step S606, it is determined whether or not the first language morpheme list 900 is over. The first language morpheme list 900 includes ten first language morphemes 901 “hi”, 902 “ha”, 706 “object oriented”, 903 “language”, 904 “ 905 “study”, 906 “shi”, 907 “te”, 908 “is”, and 909 “.” Are stored. In the first step S606, the first first language morpheme 901 “hi” in the first language morpheme list 900 is the object of processing. It is determined that it is not the end of the first language morpheme list 900. Therefore, it progresses to step S607 (step S606: No).

  In step S607, it is determined whether the unknown word 704 “object-oriented” in the unknown word list 800 and the first language morpheme 901 “hi” in the first language morpheme list 900 are the same. Since it is not the same notation, it progresses to step S608 (step S607: No).

  In step S608, it is determined whether or not the unknown word 704 “object oriented” is included in the first language morpheme 901 “he”. Since it is not included, the next first language morpheme 902 “ha” in the first language morpheme list 900 is processed, and the process returns to step S606.

  Similarly to the first language morpheme 901 “he”, the processes in and after step S606 are executed for the first language morpheme 902 “ha”. Then, the third unknown word morpheme 706 “object-oriented” in the first language morpheme list 900 is processed, and the process returns to step S606.

Similarly, the process proceeds from step S606 to step 607.
In step S607, since the unknown word 704 “object-oriented” in the unknown word list 800 and the unknown word morpheme 706 “object-oriented” in the first language morpheme list 900 are the same, the process proceeds to step S609 (step S607: Yes). ).

  In step S <b> 609, the unknown word third language extraction unit 18 performs the unknown word recognition unit 14 of the Japanese-English translation 705 “He is studying object oriented language.” Translated by the first language third language translation unit 15. The unknown word corresponding English 707 “object oriented” corresponding to the recognized unknown word 704 “object oriented” is extracted. Then, the third language notation corresponding to the unknown word morpheme 706 “object oriented” is defined as an unknown word corresponding English 707 “object oriented” corresponding to the unknown word 704 “object oriented”. This unknown word corresponding English 707 “object oriented” is stored in the third language notation list. At this time, one unknown word corresponding English 707 “object oriented” is stored in the third language notation list.

  Then, the process returns to step S606, and the same steps are performed for the first language morpheme 903 “language”, 904 “on”, 905 “study”, 906 “do”, 907 “te”, 908 “is”, and 909 “.”. The processing after S606 is executed. After performing the same processing for the first language morpheme 909 “.”, When returning to step S606, it is determined that the first language morpheme list 900 is the end. Therefore, the process returns to step S605 (step S606: Yes).

  In step S605, since the record registered in the unknown word list 800 is only the unknown word 704 “object oriented”, it is determined that the record is the end of the unknown word list 800 (step S605: Yes). Therefore, the unknown word third language extraction process ends. Steps S610 and S611 will be described later with reference to FIG.

FIG. 10 is a flowchart showing the flow of the alternative notation generation process.
In order to make the third language notation registered in the third language notation list the target of processing in step S1002 and subsequent steps in order, in step S1001, it is determined whether or not the end of the third language notation list corresponding to the unknown word. The third language notation list corresponding to the unknown word is created in step S609 in FIG. For example, in the above example used for the description of FIG. 6, one unknown word-corresponding English 707 “object oriented” is stored in the third language notation list.

  If it is determined that the third language notation list is over (step S1001: Yes), the alternative notation generation process is terminated. On the other hand, if it is determined that it is not the end of the third language notation list (step S1001: No), in step S1002, unknown words corresponding to the third language notation are sequentially acquired from the third language notation list. In the example illustrated in FIG. 7, the unknown word 704 “object oriented” corresponding to the unknown word corresponding English 707 “object oriented” is acquired.

  In step S1003, the third language morpheme analyzer 17 is used to divide the third language notation into a plurality of morphemes and create a third language notation morpheme list. In the example shown in FIG. 7, the unknown word correspondence English 707 “object oriented” is divided into the unknown word correspondence English morpheme 708 “/ object / oriented /”, and the two third language morphemes “object” and “oriented” are the first. Stored in the trilingual morpheme list.

  In order to make the third language morpheme registered in the third language notation morpheme list the target of processing in step S1005 and subsequent steps in order, in step S1004, it is determined whether or not the end of the third language notation morpheme list created in step S1003. .

  If it is determined that the third language notation morpheme list is over (step S1004: YES), the process proceeds to step S1010 described later. On the other hand, when it is determined that it is not the end of the third language notation morpheme list (step S1004: No), in step S1005, the first language third language bilingual dictionary 16 is searched with words of the third language divided into each morpheme. Then, a first language notation list is created with the searched first language notation. The first language notation list is created in the order registered in the first language third language parallel translation dictionary 16, usually in the order of frequency. In the example shown in FIG. 7, the first language third language parallel translation dictionary 16 is searched with the third language morpheme “object”, and the English morpheme-corresponding Japanese 709 “object, object, purpose” which is the searched first language notation. The first language expression list is created.

  In order to make the first language notation registered in the first language notation list the target of processing in step S1007 and subsequent steps in order, in step S1006, it is determined whether or not the end of the first language notation list created in step S1005.

で第二言語表記リストを作成する。 If it is determined that the first language notation list is over (step S1006: YES), the process returns to step S1004. On the other hand, when it is determined that it is not the end of the first language notation list (step S1006: No), in step S1007, the first language second language parallel translation dictionary 13 is searched with the first language notation, and the searched second language. Create a second language notation list with notation. In the example shown in FIG. 7, a search is performed using Japanese 709 “object, object, purpose” corresponding to English morpheme.

To create a second language list.

  In order to make the second language notation registered in the second language notation list the processing target of step S1009 in order, it is determined in step S1008 whether or not the second language notation list is an empty list.

が格納されているので、空リストではない。そして、代替語バッファには、

に対応する第一言語表記「対象」を追加する。 If it is determined that the list is an empty list (step S1008: Yes), the process returns to step S1006. If it is determined that the list is not an empty list (step S1008: No), the first language notation is added to the alternative word buffer in step S1009. Then, the process returns to step S1004. In the example shown in FIG. 7, at the beginning of the second language notation list,

Is stored, so it is not an empty list. And in the alternative word buffer,

The first language notation “object” corresponding to is added.

  Then, in the second step S1004, it is determined that it is not the end of the third language notation morpheme list (step S1004: No), and in the second step S1005, the third language morpheme “oriented” is used as the first language third language parallel translation. The dictionary 16 is searched, and a first language notation list is created with English morpheme-corresponding Japanese 710 “oriented” which is the searched first language notation.

  In the second step S1006, it is determined that it is not the end of the first language notation list (step S1006: No), and in the second step S1007, the first language second language parallel translation dictionary 13 is searched with the first language notation, A second language notation list is created with the retrieved second language notation. In the example shown in FIG. 7, the English morpheme-compatible Japanese 710 “oriented” is searched, and the searched “face-to-face” second language notation list is created.

  In the second step S1008, “face-to-face” is stored at the beginning of the second language notation list, so it is not an empty list. Then, the first language notation “oriented” corresponding to “face-to-face” is added to the alternative word buffer. At this point (second time step S1009), “subject oriented” is stored in the alternative word buffer.

  In step S1004 for the third time, it is determined that the third language notation morpheme list is over (step S1004: Yes), and in step S1010, it is added in step S1009 as an alternative notation for the unknown word acquired in step S1002. The first language notation in the alternative word buffer is output to the notation replacement table. In the example illustrated in FIG. 7, the alternative notation 711 “object-oriented” in the alternative word buffer is output to the notation replacement table as the alternative notation of the unknown word 704 “object oriented”.

  In step S1011, the substitute word buffer is cleared, and the process returns to step S1001.

FIG. 11 is a flowchart showing the flow of input sentence correction processing.
Since the alternative notations registered in the notation replacement table created in step S1010 of FIG. 10 are to be processed sequentially in step S1102 and subsequent steps, in step S1101, it is determined whether or not the end of the notation replacement table.

  When it is determined that the end of the notation replacement table is reached (step S1101: Yes), the input sentence correction process is terminated. On the other hand, if it is determined that it is not the end of the notation replacement table (step S1101: No), in step S1102, a sentence in the first language (day) input by the input unit 11 with an unknown word that is a replacement source notation is searched. . In the example shown in FIG. 7, the unknown word 704 “object-oriented” is searched for the original text 701 “he is studying object-oriented language”.

  In step S1103, the hit notation is replaced with an alternative notation. In the example shown in FIG. 7, the portion corresponding to the unknown word 704 “object-oriented” in the original text 701 “he is studying object-oriented language” is represented by the alternative notation 711 “object-oriented”. Replace. The result of the replacement is the corrected original text 712 “He is studying object-oriented language”.

と翻訳される。 When the processes described in FIGS. 6, 10, and 11 are executed, the original text 701 “He is studying the object-oriented language.” I am studying. " Then, when the first language / third language translation unit 15 translates the corrected original sentence 712 “he is studying the object-oriented language”, the corrected Japanese / Chinese translation sentence 713 is obtained.

Is translated.

  Next, an example in the case of proceeding to step S610 and step S611 in FIG. 6 will be described using FIG. 6, FIG. 12, and FIG.

となる。また、翻訳の過程で実行される形態素解析の結果として、原文形態素１２０１「/彼/は/オブジェクトオリエンテッドランゲージ/を/勉強/し/て/いる/。/」を取得する。図７に示した原文形態素７０３「/彼/は/オブジェクトオリエンテッド/ランゲージ/を/勉強/し/て/いる/。/」とは、「オブジェクトオリエンテッドランゲージ」を１つの形態素と判断するか２つの形態素と判断するかの違いである。この違いは、第一言語第三言語翻訳部１５の処理能力に起因する。 In step S601 in FIG. 6, the first language / second language translation unit 12 translates the original text 701 in FIG. 12, “He is studying the object-oriented language”. The translation result is a Japanese-Chinese translation 702

It becomes. In addition, as a result of the morphological analysis executed in the translation process, the original morpheme 1201 “/ hi / ha / object-oriented language /// study / do ////” is acquired. The textual morpheme 703 “/ he / ha / object-oriented / language /// study / do / do / is /” shown in FIG. 7 is “object-oriented language” determined as one morpheme? This is the difference between judging two morphemes. This difference is caused by the processing capability of the first language / third language translation unit 15.

と、取得した原文形態素１２０１「/彼/は/オブジェクトオリエンテッドランゲージ/を/勉強/し/て/いる/。/」に基づいて、第一言語第二言語対訳辞書１３に登録されていない第一言語（日）の表現、未知語７０４「オブジェクトオリエンテッド」を「未知語」として認識する。そして、認識した未知語を、図８に示すような未知語リスト８００に作成する。 In step S602, the unknown word recognition unit 14 translates the Japanese-Chinese translation 702.

Based on the acquired original morpheme 1201 “/ he / ha / object-oriented language /// study / do / do / is /./”, it is not registered in the first language / bilingual bilingual dictionary 13. Recognize an unknown word 704 "object oriented" as an "unknown word" in one language (day). Then, the recognized unknown word is created in an unknown word list 800 as shown in FIG.

  In step S603, the first language third language translation unit 15 translates a sentence in the same first language (day) as the sentence in which the first language is translated into the second language in step S601, and obtains a morphological analysis result. . In other words, the first language third language translation unit 15 translates the sentence 701 “he is studying object-oriented language”, which is the first language (day) sentence input by the input unit 11. The translation result is a Japanese-English translation 705 “He is studying object oriented language.” In addition, as a result of the morphological analysis executed in the translation process, the original morpheme 1201 “/ hi / ha / object-oriented language /// study / do ////” is acquired.

  In step S604, as a result of the morphological analysis performed by the first language / second language translation unit 12 in step S601, the original morpheme 1201 “/ he / has / object-oriented language // study / do / do / is /./ The first language morpheme list 1300 as shown in FIG.

  Then, in order to make the unknown words registered in the unknown word list 800 the target of processing in step S606 and subsequent steps, in step S605, it is determined whether or not the end of the unknown word list 800 is reached. In the unknown word list 800, as shown in FIG. 8, one unknown word 704 "object oriented" is stored. In the first step S605, the first unknown word 704 “object-oriented” in the unknown word list 800 is a processing target. It is determined that it is not the end of the unknown word list 800. Therefore, it progresses to step S606 (step S605: No).

  In step S606, it is determined whether the first language morpheme list 1300 is over. In the first language morpheme list 1300, as shown in FIG. 13, nine first language morphemes 1301 "he", 1302 "ha", 1202 "object-oriented language", 1303 "on", and 1304 "study" 1305 “do”, 1306 “te”, 1307 “is”, and 1308 “.” Are stored. In the first step S606, the first first language morpheme 1301 “hi” in the first language morpheme list 1300 is the target of processing. It is determined that it is not the end of the first language morpheme list 1300. Therefore, it progresses to step S607 (step S606: No).

  In step S607, it is determined whether the unknown word 704 “object-oriented” in the unknown word list 800 and the first language morpheme 1301 “he” in the first language morpheme list 1300 are the same. Since it is not the same notation, it progresses to step S608 (step S607: No).

  In step S608, it is determined whether or not the unknown word 704 “object oriented” is included in the first language morpheme 1301 “he”. Since it is not included, the next first language morpheme 1302 “ha” in the first language morpheme list 1300 is processed, and the process returns to step S606.

  Similar to the first language morpheme 1301 “he”, the process from step S606 onward is executed for the first language morpheme 1302 “ha”. Then, the third unknown word morpheme 1202 “object-oriented language” in the first language morpheme list 1300 is processed, and the process returns to step S606.

Similarly, the process proceeds from step S606 to step 607.
In step S607, since the unknown word 704 “object-oriented” in the unknown word list 800 and the unknown word morpheme 1202 “object-oriented language” in the first language morpheme list 1300 are not the same, the process proceeds to step S608 (step S607: No). ).

  In step S608, it is determined whether the unknown word is included in the morpheme. The unknown word 704 “object-oriented” is included in the unknown word morpheme 1202 “object-oriented language”. That is, unknown word 704 “object-oriented” and unknown word morpheme 1202 “object-oriented language” are not identical (step S607: No), and unknown word 704 “object-oriented” is unknown word morpheme 1202 “object-oriented language”. (Step S608: Yes), in step S610, the corresponding third language (English) other than the included portion is searched. Since “Object Oriented” is included, “Language” is included except for the included portion, and the corresponding third language (English) “language” is searched.

  In step S611, the third language (English) “language” of the search part searched in step S610 is removed from the unknown word morpheme-corresponding English 1203 “object oriented language” corresponding to the unknown word morpheme 1202 “object-oriented language”. The remaining unknown word corresponding English 1204 “object oriented” is set as a third language notation corresponding to the unknown word 704 “object oriented”. This unknown word corresponding English 1204 “object oriented” is stored in the third language notation list. At this time, one unknown word corresponding English 1204 “object oriented” is stored in the third language notation list.

  Then, the process returns to step S606, and the processes in and after step S606 are similarly performed for the first language morpheme 1303 “to”, 1304 “study”, 1305 “do”, 1306 “te”, 1307 “is”, and 1308 “.”. Run. After the same processing is executed for the first language morpheme 1308 “.”, When the process returns to step S606, it is determined that the first language morpheme list 1300 is over. Therefore, the process returns to step S605 (step S606: Yes).

  In step S605, since the record registered in the unknown word list 800 is only the unknown word 704 “object oriented”, it is determined that the record is the end of the unknown word list 800 (step S605: Yes). Therefore, the unknown word third language extraction process ends.

  The presence of unknown words for machine translation is a major factor that significantly degrades translation quality. In particular, the problem of unknown words in a translation system between minor languages with a small number of dictionary registration words is serious. The machine translation process makes it possible to reduce the occurrence frequency of unknown words, greatly improving machine translation quality.

Next, a second embodiment will be described with reference to FIG.
FIG. 14 is a diagram illustrating a configuration of the machine translation apparatus 2 according to the second embodiment.

  In FIG. 14, the machine translation device 2 includes an input unit 11, a first language / second language translation unit 12, a first language / second language parallel translation dictionary 13, and an unknown word recognition that are included in the machine translation device 1 according to the first embodiment. Unit 14, first language third language translation unit 15, first language third language parallel translation dictionary 16, third language morpheme analysis unit 17, unknown language third language segmentation unit 18, third language first language dictionary lookup unit 19 In addition to the alternative notation generation unit 20, the input sentence correction unit 21, and the output unit 22, a candidate ranking unit 141 is provided.

  The candidate ranking unit 141 ranks alternative representations of unknown words generated by the alternative representation generation unit 20 according to a predetermined priority. For example, the priority of the alternative notation with high use frequency is increased, and the priority of the alternative notation with low use frequency is decreased. Alternatively, a large-scale corpus such as the World Wide Web connected via an external network such as the Internet may be searched using alternative notation, and ranked by the number of hits.

  Then, the input sentence correction unit 21 substitutes the unknown words recognized by the unknown word recognition unit 14 among the sentences of the first language (day) input by the input unit 11 by the candidate ranking unit 141. The sentence of the input first language (day) is corrected by replacing it with an alternative notation having a high priority.

  As a result, even if the alternative notation candidate generated by the alternative notation generation unit 20 is an expression that is rarely read and written, an expression that is frequently used daily is given priority over an unusual expression, and the translation result is also a general term. Translation quality increases by being close to.

Next, a third embodiment will be described with reference to FIG.
FIG. 15 is a diagram illustrating a configuration of the machine translation apparatus 3 according to the third embodiment.

  In FIG. 15, the machine translation device 3 includes an input unit 11, a first language second language translation unit 12, a first language second language parallel translation dictionary 13, an unknown word recognition unit 14, and a plurality of first language third language translation units. 15A to 15C, a plurality of first language third language parallel translation dictionaries 16A to 16C, a plurality of third language morpheme analyzers 17A to 17C, an unknown word third language segmentation unit 18, and a plurality of third language first language dictionary lookup units 19A to 19C, an alternative notation generation unit 20, an input sentence correction unit 21, and an output unit 22.

  Input unit 11, first language second language translation unit 12, first language second language parallel translation dictionary 13, unknown word recognition unit 14, unknown word third language segmentation unit 18, alternative notation generation unit 20, input sentence correction unit 21 The output unit 22 is the same as the input unit 11 and the like included in the machine translation apparatus 1 according to the first embodiment.

  The plurality of first language / third language translation units 15A to 15C perform morphological analysis on the sentence of the first language (day) input by the input unit 11, and select one of the first language / third language parallel translation dictionaries 16A to 16C. Referring to FIG. 4, translation is made into a sentence in a third language (English) different from the first language (Japanese) and the second language (middle), but the morphological analysis method or the translation method is different.

  In the plurality of first language third language parallel translation dictionaries 16A to 16C, at least one expression of the third language (English) is registered in association with at least one expression of the first language (Japanese). , Content is different.

  The plurality of third language morpheme analyzers 17A to 17C divide the third language (English) sentence into a plurality of morphemes, but the morpheme analysis methods are different.

  The plurality of third language first language dictionary lookup units 19A to 19C refer to any one of the first language third language parallel translation dictionaries 16A to 16C, and correspond to the first language (English) corresponding to the word of the third language (English) ( Japanese), but the translation method is different.

  As a result, the occurrence frequency of unknown words can be further reduced, and the quality of machine translation is greatly improved.

Next, a fourth embodiment will be described with reference to FIG.
FIG. 16 is a diagram illustrating the configuration of the fourth embodiment.

  In FIG. 16, the machine translation device 1 is connected to a terminal device 1601A and a terminal device 1601B via a communication network 1600 such as the Internet.

  With this configuration, the sentence in the first language to be translated is input from the terminal device 1601A or the terminal device 1601B via the communication network 1600 to the input unit 11 included in the machine translation device 1. Then, the text in the second language translated by the first language / second language translation unit 12 included in the machine translation device 1 is output to the terminal device 1601A or the terminal device 1601B via the communication network 1600.

  Thereby, anyone can use the machine translation apparatus 1 via the communication network 1600 such as the Internet.

  Instead of the machine translation device 1, the machine translation device 2 or the machine translation device 3 may be connected to the terminal device 1601A and the terminal device 1601B via a communication network 1600 such as the Internet.

Next, a fifth embodiment will be described with reference to FIG.
FIG. 17 is a diagram illustrating the configuration of the fifth embodiment.

  In FIG. 17, the first language / second language translation unit 12, the first language / second language parallel translation dictionary 13, the first language / third language translation unit 15, the first language / third language parallel translation dictionary 16, and the machine translation device 1 are provided. The third language morpheme analysis unit 17 or the third language first language dictionary lookup unit 19 is incorporated from the outside via a communication network 1600 such as the Internet.

Next, a sixth embodiment will be described with reference to FIG.
FIG. 18 is a diagram illustrating a configuration of the machine translation apparatus 4 according to the sixth embodiment.

  18, the machine translation device 4 includes a first language / second language translation unit 12, a first language / second language parallel translation dictionary 13, a first language / third language translation unit 15, and a first language / third language parallel translation. Instead of the dictionary 16, the third language morpheme analysis unit 17, or the third language first language dictionary lookup unit 19, an external one connected by a communication network 1600 such as the Internet is used.

  Although the embodiments of the present invention have been described above, the above-described machine translation apparatuses 1, 2, 3, and 4 can be realized using a normal information processing apparatus (computer). That is, the machine translation devices 1, 2, 3, and 4 include a CPU, a memory such as a RAM and a ROM, an input device, an output device, an external recording device, a medium driving device, and a network connection device. These are connected to each other by a bus.

  The memory stores programs and data used for the machine translation apparatuses 1, 2, 3, and 4. The CPU executes the above-described machine translation process by executing the program using the memory.

  The input device is, for example, a keyboard, a pointing device, or the like, and is used for inputting instructions and information from the user. The output device is, for example, a display device, a printer, a speaker, or the like, and is used for outputting an inquiry to a user and a processing result.

  The external recording device is, for example, a magnetic disk device, an optical disk device, a magneto-optical disk device, a tape device, or the like. The external recording device also includes a hard disk drive. The machine translation apparatuses 1, 2, 3, and 4 can store programs and data in the external recording apparatus and use them by loading them into a memory.

  The medium driving device drives a portable recording medium and accesses the recorded contents. The portable recording medium is a memory device, a flexible disk, an optical disk, a magneto-optical disk, or the like. The portable recording medium includes a CD-ROM (Compact Disk Read Only Memory), a DVD (Digital Versatile Disk), a USB (Universal Serial Bus) memory, and the like. The user can store the program and data in this portable recording medium and load them into the memory for use.

  As described above, the computer-readable recording medium for storing the program and data used in the machine translation process includes physical (non-temporary) recording such as a memory, an external recording device, and a portable recording medium. Media included.

  The network connection device is a communication interface that is connected to a wired or wireless communication network such as a LAN (Local Area Network) and performs data conversion accompanying communication. The machine translation devices 1, 2, 3, 4 can receive programs and data from an external device via a network connection device, and use them by loading them into a memory.

  For example, programs and data stored in the external recording device are loaded into the memories of the machine translation devices 1, 2, 3, and 4. In addition, an external device that can be connected via a network connection device generates a carrier signal for carrying a program or data, and transmits it to a machine translation device 1, 2, 3, 4 via any transmission medium on the communication network. To do.

  Although the disclosed embodiments and their advantages have been described in detail, those skilled in the art can make various modifications, additions and omissions without departing from the scope of the present invention as explicitly set forth in the claims.

1, 2, 3, 4 Machine translation device 11 Input unit 12 First language second language translation unit 13 First language second language parallel translation dictionary 14 Unknown word recognition unit 15, 15A, 15B, 15C First language third language translation Part 16, 16A, 16B, 16C First language third language parallel translation dictionary 17, 17A, 17B, 17C Third language morphological analysis part 18 Unknown word third language extraction part 19, 19A, 19B, 19C Third language first language Dictionary lookup unit 20 Alternative notation generation unit 21 Input sentence correction unit 22 Output unit 141 Candidate ranking part 701 Original sentence 702 Japanese-Chinese translation 703, 1201 Original sentence morpheme 704 Unknown word 705 Japanese-English translation sentence 706, 1202 Unknown word morpheme 707, 1204 Unknown English corresponding to word 708 English morpheme corresponding to unknown word 709, 710 Japanese corresponding to English morpheme 711 Alternative notation 712 Modified original text 13 Modified Japanese-Chinese translation 800 Unknown word list 900, 1300 First language morpheme list 901, 902, 903, 904, 905, 906, 907, 908, 909, 1301, 1302, 1303, 1304, 1305, 1306, 1307, 1308 1st language morpheme 1203 English for unknown word morpheme 1600 Communication network 1601A, 1601B Terminal device

Claims (8)

An input means for inputting a sentence in the first language to be translated;
A first-language second-language bilingual dictionary in which at least one second-language expression is registered in association with at least one expression of the first language;
Morphological analysis is performed on the sentence in the first language input by the input means, and the sentence is translated into a sentence in a second language for translation purposes different from the first language by referring to the first language / second language parallel translation dictionary. A first language second language translation means;
An unknown word recognizing means for recognizing an unknown word in which the representation of the second language is not registered in the first language bilingual bilingual dictionary based on the translation result and the morphological analysis result by the first language second language translating means When,
A first language third language bilingual dictionary in which at least one third language expression different from the first language and the second language is registered in association with at least one expression of the first language;
A first language and third language translation means for performing morphological analysis on the sentence of the first language input by the input means, and referring to the first language and third language parallel translation dictionary to translate the sentence into the third language; ,
Of the sentence in the third language translated by the first language and third language translation means, an unknown word third language cutting means for cutting out a third language corresponding to the unknown word recognized by the unknown word recognition means;
Third language morpheme analyzing means for dividing the third language extracted by the unknown word third language extracting means into a plurality of morphemes;
A third language first language dictionary lookup means for searching for words in the first language corresponding to the divided morphemes with reference to the first language third language parallel translation dictionary;
Substitution of the unknown word recognized by the unknown word recognition means by combining a plurality of words of the first language corresponding to the plurality of divided morphemes using the third language first language dictionary lookup means An alternative notation generation means for generating a notation;
Of the sentences in the first language input by the input unit, the unknown word recognized by the unknown word recognition unit is replaced with the alternative notation generated by the alternative notation generation unit, whereby the input first An input sentence correcting means for correcting sentences in one language;
With
The machine translation apparatus characterized in that the first language / second language translation means translates the text in the first language corrected by the input sentence correction means into a text in the second language. Candidate ranking means for ranking the alternative notation of the unknown word generated by the alternative notation generating means according to a predetermined priority,
The input sentence correction means has priority among the alternative notations ranked by the candidate ranking means for the unknown words recognized by the unknown word recognition means among the sentences in the first language input by the input means. The machine translation apparatus according to claim 1, wherein the sentence in the first language input by the input unit is corrected by replacing it with an alternative notation having a high degree.   The machine translation apparatus according to claim 1 or 2, wherein the first language / third language parallel translation dictionary includes a plurality of dictionaries. From the terminal device via the communication network, input the sentence in the first language into the input means,
Outputting the sentence in the second language translated by the first language / second language translation means to the terminal device via the communication network;
The machine translation apparatus according to claim 1, wherein the machine translation apparatus is a machine translation apparatus.   The first language second language translation means, the first language second language parallel translation dictionary, the first language third language translation means, the first language third language parallel translation dictionary, the third language morpheme analysis means, or the The machine translation apparatus according to any one of claims 1 to 4, wherein the third language first language dictionary lookup unit is incorporated from outside via a communication network.   The first language second language translation means, the first language second language parallel translation dictionary, the first language third language translation means, the first language third language parallel translation dictionary, the third language morpheme analysis means, or the The machine translation device according to any one of claims 1 to 4, wherein the third language first language dictionary lookup unit is installed outside connected by a communication network. In the machine translation device computer,
Input the first language sentence to be translated,
A morphological analysis is performed on the input sentence in the first language, and at least one second language expression is registered in association with at least one expression in the first language. Refer to the dictionary and translate it into a second language sentence for translation purposes different from the first language,
Based on the translation result and the morphological analysis result from the first language to the second language translation, the unknown language in which the second language expression is not registered in the first language second language parallel translation dictionary is recognized,
Morphological analysis is performed on the inputted sentence in the first language, and at least one third language expression different from the first language and the second language is associated with at least one expression in the first language. Refer to the registered first language third language parallel translation dictionary, and translate it into the third language sentence,
Of the sentences in the third language translated by translation from the first language to the third language, the third language corresponding to the recognized unknown word is cut out,
Dividing the extracted third language into a plurality of morphemes;
With reference to the first language third language parallel dictionary, the words of the first language corresponding to the divided morphemes are searched,
By combining a plurality of words in the first language corresponding to the plurality of divided morphemes, generating an alternative notation of the recognized unknown word,
Of the inputted sentence in the first language, by replacing the recognized unknown word with the generated alternative notation, to correct the inputted sentence in the first language,
Translating the modified first language sentence into a second language sentence;
A machine translation program characterized by that. In the machine translation method,
Machine translation device
Enter the first language text to be translated,
A morphological analysis is performed on the input sentence in the first language, and at least one second language expression is registered in association with at least one expression in the first language. Refer to the dictionary and translate it into a second language sentence for translation purposes different from the first language,
Based on the translation result from the first language to the second language translation and the morphological analysis result, an unknown word in which the expression of the second language is not registered in the first language bilingual bilingual dictionary is recognized.
Morphological analysis is performed on the inputted sentence in the first language, and at least one third language expression different from the first language and the second language is associated with at least one expression in the first language. Referring to the first language third language parallel translation dictionary registered in the
Of the third language sentence translated by translation from the first language to the third language, cut out a third language corresponding to the recognized unknown word,
Dividing the extracted third language into a plurality of morphemes;
Referencing the first language third language parallel dictionary and searching for words in the first language corresponding to the divided morphemes,
Generating an alternative notation of the recognized unknown word by combining a plurality of words of the first language corresponding to the plurality of divided morphemes;
Of the inputted sentence in the first language, by replacing the recognized unknown word with the generated alternative notation, to correct the inputted sentence in the first language,
Translating the modified first language sentence into a second language sentence;
A machine translation method characterized by the above.

Images