JP3323517B2 - Machine translation equipment - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation apparatus, and more particularly, to a machine translation apparatus which facilitates post-processing in machine translation.

[0002]

2. Description of the Related Art For example, a machine translator for mechanically translating a Japanese sentence into English has been put to practical use. This machine translation device is effective in translating documents such as manuals and magazines, for example, by selecting unified translations using dictionaries such as a basic dictionary, a field dictionary, and a user dictionary. That is, when a translator translates, different translations and expressions may be made in the file based on the mood and sentence of the sentence for the same original language expression even in the same translator. Such translations can be difficult or misleading for third parties. Machine translators do not have this problem because they select unified translations.
This is particularly effective in that post-processing for unifying translated words is not required.

[0003]

By the way, even in a translation office having such a machine translation apparatus, if a large number of translations are performed within a fixed period, translators who are geographically dispersed and operate. In contrast, the current situation is to use shared translation and a wide area network to increase the processing amount, and to quickly distribute and collect manuscripts and translation results. In other words, if a single customer requests translation of a large number of manuals and other documents for several volumes with a fixed delivery date, the translation office will complete all translation work within the delivery date. The original text is appropriately divided and assigned to multiple translators for translation.

However, since a machine translation device generally uses a plurality of translated words and a plurality of notations for the same original language expression, the selection depends on the qualities of each translator such as the level of understanding of the original language and the inferior knowledge of the original sentence. It depends greatly. As a result, when translating a large number of documents in a shared manner, even if each translator translates using the same machine translator and the same dictionary, if the shared translation results are compiled, There is a problem that considerable inconsistency of translated words occurs. In particular, inconsistencies in translated words occur more frequently due to differences in notations due to foreign words, technical terms, or customer habits.

[0005] For this reason, when each translator finishes translating the assigned original text by the machine translator, the translators collect the translated text, perform a sufficient inspection such as unification of the translated words, and deliver the entire translated text to the customer. ing. Therefore, it takes considerable time and effort to unify the translated words of each translator. Accordingly, a first object of the present invention is to provide a machine translation apparatus that can easily detect and confirm how uniform a plurality of translated words and expressions translated using one original language expression as an axis. It is in. A second object of the present invention is to provide a machine translation apparatus capable of efficiently correcting and unifying detected inconsistent translations.

[0006]

According to the first aspect of the present invention, word extraction means for extracting words from an original sentence, and translation information necessary for translation such as headwords of the original language, parts of speech classification, and translations of the target language are stored. , A dictionary extracting means for extracting translation information stored in the dictionary from words extracted by the word extracting means, and a syntax analyzing means for analyzing a syntax of an original sentence from the translation information extracted by the dictionary extracting means. Conversion / sentence generation means for converting the syntax analyzed by the syntax analysis means and the translation information obtained by the dictionary lookup means into a target language and generating a translated sentence; and a word extracted by the word extraction means. And part- of- speech classification for the usage used in the original sentence, and the translated sentence generated by the conversion / sentence generating means.
Was a translation made to correspond, when part of speech classification and translation are the same, their
In addition the frequency using the translation of the translation correspondence list creating means for creating a translation corresponding list is immediately Bei by achieving the first object to. According to the second aspect of the invention, the first aspect is described.
Machine translation apparatus, the translated word correspondence list creating means
Indicates the features of the original sentence in which the word on the translation correspondence list exists.
The first object is achieved by creating a translated word correspondence list to which file information is added . In the invention of claim 3, wherein a plurality
Each machine created by the machine translation apparatus according to claim 1 or 2.
Add the name of the person who created the translation to the translation
Merge these translated word correspondence lists and sort them by headword
List merging means to create the entire translation list
By Bei achieve the first object.

According to the fourth aspect of the present invention, the first to third aspects are provided.
A machine translation apparatus according to any one of the inconsistencies word searching means single word and part of speech classification to search for translation is different words at the same from the translation correspondence list, the translation generated by the conversion-sentence generation means Display means for detecting and sequentially displaying a translation corresponding to the word searched by the unified word search means, and correction means for sequentially correcting a translation in the translation corresponding to the word displayed on the display means It is provided with a door
To achieve the second object Te. According to a fifth aspect of the present invention, in the machine translation apparatus according to the fourth aspect, the display means
The column uses the word searched by the unified word search means.
Including the text before and after the original and translated text
Thereby to achieve the second object by. According to a sixth aspect of the present invention, in the machine translation apparatus of the fifth aspect, the display means
The step distinguishes the word in the original and the translation from other words.
It is displayed Te to achieve the second object by. According to a seventh aspect of the present invention, there is provided the machine according to any one of the first to third aspects.
In the translator, the word and part-of-speech classification are the same but the translated words are different
Unified word searcher to search for words from translation list
Dan and the translated word of the word searched by this unified word search means
Display means for displaying the correspondence list and display on the display means
The conversion / correspondence corresponding to a specified word of the
Correct the translated words of the translation generated by the sentence generation means at once
The second object is achieved by providing a correction means .

[0008]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of a machine translation apparatus according to the present invention will be described below in detail with reference to FIGS. FIG.
Represents the configuration of the machine translation device. In FIG. 1, reference numeral 1 denotes a word extracting device for extracting words from an original sentence file 2 storing original sentences to be translated. Here, the original text file 2 is a document created by a text creation device such as a word processor or a personal computer, or a document recognized by a character recognition device.

Reference numeral 3 denotes a translation dictionary having translation information such as the headword of the original language, its part-of-speech classification, the translation of the target language, and grammatical information. It is adapted to be drawn by the dictionary pull-out device 4. Reference numeral 5 denotes a syntax analysis device that analyzes the syntax of the original sentence after the translation information is extracted, and 6 denotes a conversion / sentence generation device that converts the original sentence into a target language according to the translation information and the syntax and generates a translated sentence. The translation generated by the conversion / sentence generation device 6 is stored in the translation file 7. In the translation stored in this translation file, the original text and the translation are usually stored in a text-to-text translation, but depending on the set mode, only the translation is stored. 8 is a translation word generated by the conversion / sentence generation device 6 and finally stored in the translation file 7;
This is a translated word correspondence list creation device that creates a translated word correspondence list 9 using words obtained by cutting out the parts of speech and the like from the original text as headwords. This translated word correspondence list creating device 8 merges (merges) the translated word correspondence list 9 created by another machine translation device.
Then, an entire translation list is created.

Reference numeral 10 denotes a display device on which the translated sentence or the translated word generated by the conversion / sentence generating device 6 is displayed together with the original sentence. 11
Is an editing device having a post-editing function in which a user instructs selection or correction of a required translated word based on a display on a display device. Reference numeral 12 denotes a search / correction device which opens the file of the translated word correspondence list 9, searches for words having the same headword and different parts of speech and translated words, displays the corresponding original text and translated text on the display device 10, and displays the corresponding original text and translated text from the editing device 11. The translation of the translation file 7 is corrected according to the input.
In addition, the search and correction device 12 displays a portion corresponding to the searched headword with, for example, a display blink, a different color, an underline, or the like, and displays it differently from other portions.

The translation operation of the machine translation apparatus having the above-described configuration will be described below. Generally, a machine translation system using a syntax conversion method reads a source file and sequentially follows phases of morphological analysis, syntax analysis, conversion, and sentence generation.
Create a translation file. In the present embodiment, in the morphological analysis phase, words, symbols, and the like are separated from the original sentence file, and a dictionary is looked up using the extracted words as headwords. The word extracting device 1 extracts only words from the original sentence file 2. Then, the extracted words the headword, after subtracting the dictionary 3 in the dictionary Technology of device 4 analyzes the syntax parsing apparatus 5 performs conversion and sentence generation by the conversion-generation apparatus 6. The translation or translated word generated by the conversion / sentence generation device 6 is selected by the user by the editing function of the search / correction device 12, and the final translated text in sentence units is stored in the translated text file 7 together with the original text as a parallel translation. On the other hand, a record is created by associating the translated word finally adopted in the translated text file with the part of speech used at that time with the headword. A translated word correspondence list (file) 9 is created in accordance with all the records thus operated.
This translated word correspondence list 9 is created separately from the translated sentence file 7.

Next, the translation by the machine translation apparatus and the creation of the translation correspondence list will be described based on English-Japanese translation examples. FIG.
Shows an example of an original sentence received from a customer and an assignment to each translator performed by a translation office. The original text is taken from the manual of the printer, and it can be seen that Volume-3 is a series of documents in content. Each of the translators 1 to 3 creates a translated sentence file 7 using a machine translator according to the assignment. At this time, the translated word correspondence list 9 is created in the following procedure. First, the original sentence is separated into words by the word extraction device 1 as follows. / Installing // th
e // Toner // Cartridge and headwords such as "Installing", "the",
You get the four words "Toner" and "Carridge". Pull all dictionary 3 in the dictionary Technology of device 4 for each of these entry word, obtain a translated word as shown in Figure 3A.
Each entry word, its part of speech classification, conjugations, syntax type, etc. are also together description, parsing instrumentation <br/> location according to the contents creates a syntax tree, in one of the original, single Create a complete tree that will be the root of

Here, the complete tree is obtained by subjecting the translation order such as the position of the subject, predicate, and adverbial phrase to syntax conversion in accordance with the target language specification. A translation is selected, and a translation is generated according to the syntax tree of the target language. As a result, "Install the toner cartridge"
To get the translation. Thereafter, the translator performs post-editing by using a translation / translation word selection function of a post-editing device (not shown). Now, the translator has determined that “toner cartridge” is a technical term, and the original language expression “Toner Cartridge”
e ”is considered appropriate, and“ Toner Cartr
It is assumed that “attach an id” is selected as the final translation. At this time, since the user has not performed free editing and has obtained the final translation from the pre / post editing function provided in the machine translation device, the syntax tree as internal information has not been destroyed. Then, word elements corresponding to the original sentence and the translated sentence are extracted, and a translated word correspondence list file shown in FIG. 3B is created.

FIG. 4 shows a sorted translation correspondence list file created for the entire document handled by the translator 1 by the same operation, and FIG. 5 shows the translation by the translator 1 at this time. When looking at FIG. 4 sequentially from the top, it is easy to find a case where the translated word or the notation differs for the same combination of the headword and the part of speech. In other words, "Cart
The translation of “ridge” is “Cartridge”, and the translation of “cartridge” starting with a lowercase letter is “cartridge”. "Toner" also starts with capital letters
Translation of "toner" with lowercase "toner"
Is "toner". By the above operation, the variation of the translated word or the notation within the same translator can be easily inspected, and the character string to be corrected is known, so that the correction operation is accurate and easy.

The same operation can be performed by the translators 2 and 3. FIG. 6A shows a translated word correspondence list for the assigned translator 2, and FIG. 6B shows a translated word correspondence list of the translator 3. 7 and 8 show translations by translators 2 and 3. FIG. As described above, the translated sentence file 7 and the translated word correspondence list 9 (FIGS. 4, 6A, and 6B) created by the translators sharing the respective machine translators are stored in the ISDN (service) to which each machine translator is connected. It is collected by a machine translation device located in a translation office or the like by a wide area network such as an integrated digital network or a personal computer communication network. The translated word correspondence lists 9 collected in one machine translation device are sorted and processed collectively by the translated word correspondence list creating device 8 of the machine translation device, and the entire translated word correspondence list as shown in FIG. Is created. Further, translations for each translator (FIGS. 5, 7, and 8) are also summarized as shown in FIG.

When looking at FIG . 9 in order from the top, the unification of the translated words of the entire translated sentence can be seen at a glance. That is, in terms of headword and part-of-speech pairs that appear a plurality of times between different persons in charge, unification of translated words can be classified into three as shown in FIGS. 11, 12, and 13. Figure 11 is intended to be confirmed as being unified to the same translation, 12 are those notation of translation is different depending on the representation of the original. FIG. 13 shows that although the notation of the original text is the same, there are a plurality of different translations in the dictionary as shown in FIG. 14, and each translator uses the post-editing function of another translation / different translation. Have different translations. The user can unify the translated words of the translated sentence by taking the following measures for the above three classifications, and can improve the quality of translation. That is, in the case of FIG. 12, it is reconsidered whether the difference in the notation of the original sentence should be reflected in the difference in the translated word. In the case of FIG. 13, since the notation of the original text is the same, the translated words are unified unless it is unnatural.

Next, the operation of correcting and unifying untranslated translated words will be described. The search / correction device 12 horizontally translates the original sentence and the translated sentence of the translated sentence file 7 received from each translator as shown in FIG. In the machine translation apparatus of the present embodiment, there are a dialogue correction mode in which the user interactively corrects and a batch correction mode in which the parts where the translated words are inconsistent are corrected collectively. The editing device 11 sets the mode. Now, it is assumed that the interactive correction mode has been set, and the operation thereof will be described with reference to FIG.

First, the search / correction device 12 checks whether or not there are any remaining headwords in the entire translated word correspondence list (FIG. 9). Is checked whether they are in the same line (step 2). If the headwords in the adjacent rows are not the same (step 2; N), the process proceeds to the next headword (step 3) and step 1
Return to On the other hand, when the same line of the headword is found, the same case is assumed over three or more lines, and the same procedure is repeated to identify a range recognized as the same headword (step 2; Y). In the range, it is checked whether or not the part of speech is the same (step 4). Here, if the part of speech is the same (step 4; Y), it is next checked whether or not the translation is the same, and if the translation is the same (step 5; Y). ), Even if the shared translation is made, it is found that the same translated word is unified, and the process proceeds to the next headword (step 3). Examples of such headwords are “from”, “it”,
“Its”, “the”, and “then”.

If the translated word is selected differently in spite of the same part of speech (step 5; N), FIG. 15 is used based on the pair of the file name and the person in charge of the line in FIG. , The file in which the word is used is uniquely identified, and the file is opened (step 6). After the file is opened, the location where the word is used is searched, and if there is a corresponding location (step 7; Y), the original sentence and the translated sentence are displayed on the display device 10 in units of sentences (step 8). When displaying a sentence bilingual display device 10, the search-correcting device 12, the corresponding words of the original and translation <br/> statement to distinguish it from the other parts, an instruction to display the blink.

Here, the operator operates the editing device 11 while referring to the bilingual sentence displayed on the display device 10, and interactively inputs an instruction for a necessary correction. The search and correction device 12 converts the translated file 7 displayed on the display device 10 according to the correction instruction supplied from the editing device 11.
Is corrected (step 9). After the correction is completed, the process returns to step 7 to proceed to the next applicable portion, and steps 8 and 9 are repeated until there is no applicable portion, to perform successive correction. If the last applicable point is completed (Step 7;
N), proceed to the next headword (step 3). As an example of this, the translated words “Carridge” and “cartridge”, “Magazine” and “magazine”, “ton”
As er "and of" Toner ", the difference on the notation is reflected in the translation, and what has become a different translation," re
As in the translation of "move", there are also those that differ depending on the variety of the corresponding Japanese expressions, such as "remove", "remove", and "remove".

On the other hand, if it is determined in step 4 that the part of speech is different for the same headword (N), the process proceeds to step 6 and the same operation is performed in order to confirm whether or not the translation is good from the context of the sentence. For example, a headword given as an example in this case corresponds to a case where a part of speech having a different verb or adjective is used in the original sentence in “Installing” in FIG. In the above description, a method in which the user interactively makes corrections has been specifically described. However, when making corrections collectively for different headwords, the operator specifies the batch correction mode using the editing device 11. Then, based on the translated word correspondence list 9 in FIG. 9, the translation / notation to be unified is uniquely designated. The search / correction device 12 corrects the batch by automatically assigning the translated word specified by the editing device 11 to the corresponding portion of the translated word correspondence list 9. In addition,
The translation may be corrected by using another translation function (not shown) provided in the machine translation apparatus, or a function such as a character string search and conversion of a Japanese word processor.

In the above-described embodiment, the operation has been described based on the English-Japanese translation example. However, by performing the same configuration and operation, the language pair is not limited to English-Japanese and machine translation can be applied. It can be applied to every possible language pair.

[0023]

According to the present invention as described above,
The following effects can be achieved. In the original text, when a headword and the same headword are present a plurality of times, it is possible to easily confirm whether the translated word and the notation given by the translator are unified in each case. In addition, since it is possible to quantitatively confirm whether the same translation is used for the same headword and usage in the file translated by each translator, it is possible to estimate the amount of correction work in advance and to unify translations. The amount and ratio can be determined objectively. Ma
Also occurs when translations are shared by multiple machine translation devices.
The translation and notation across different translators.
It can be easily checked and the quality of the translation can be improved. Also, if each translator creates a translation across multiple files,
There is no need to create a translation correspondence list for each file, and translation unification between multiple files can be easily checked.

Also, based on the entire translated word correspondence list, the inconsistency of the translated words of the entire document when translated by a plurality of translators or the same translator in different situations is determined.
Since display and correction are sequentially performed on the display device, inconsistencies in translated words and differences in notation can be detected and corrected without fail throughout the document. In addition , since the original sentence of the word searched by the unified word search means and the sentence before and after that used in the translated sentence are displayed, when a different translated word is selected in the translated sentence, the context of the corresponding part is considered. Unnecessary corrections can be made by concentrating on the unification of the translations, and unnatural translations can be prevented, thereby preventing the overall translation quality from deteriorating. In addition , since the corresponding portion is associated with the original text and is displayed separately from the preceding and following words , the operator can quickly recognize the corresponding portion. In particular , when the sentence is a long sentence or when there are a plurality of corresponding portions in one sentence, the correction can be performed without fail and the correction time can be reduced. Further, when the unstructured translation is found, the user only needs to specify the translation-out unified all, can be carried out collectively corrected throughout translation, it is possible to shorten the correction time.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of a machine translation apparatus according to the present invention.

FIG. 2 is a diagram showing the contents of an original sentence which is shared and translated by a machine translation device.

[3] A is a diagram illustrating a headword, each translated word with the headword, B is a diagram showing a translation and part of speech and the selected headword.

FIG. 4 is a diagram showing a translated word correspondence list created by the translator 1;

FIG. 5 is a diagram showing a bilingual sentence created by a translator 1;

6A is a diagram showing a translated word correspondence list created by the translator 2; FIG. 6B is a diagram showing a translated word correspondence list created by the translator 3;

FIG. 7 is a diagram showing a bilingual sentence created by a translator 2;

FIG. 8 is a diagram showing a bilingual sentence created by a translator 3;

9 is a diagram showing the overall translation-enabled list.

FIG. 10 is a diagram showing an entire bilingual sentence.

FIG. 11 is an explanatory diagram showing ones that can be confirmed from the entire translated word correspondence list to be unified into the same translated word.

FIG. 12 is an explanatory diagram showing an example in which a translated word is different from the entire translated word correspondence list due to a difference in original sentence notation.

FIG. 13 is an explanatory diagram showing, from the whole translated word correspondence list, those having the same original sentence but different translated words.

FIG. 14 is an explanatory diagram showing an example of a word translation.

FIG. 15 is an explanatory diagram showing a state in which a translated sentence file coming from each translator is horizontally configured during the unification operation of translated words.

FIG. 16 is a flowchart of an operation of correcting and unifying non-unified translated words interactively and sequentially.

[Explanation of symbols]

1 word cut-out device 2 original file 3 dictionary 4 dictionary pull-out device 5 syntax analysis apparatus 6 conversion and generation apparatus 7 translation file 8 translation corresponding list creating device 9 translation correspondence list 10 display device 11 editing apparatus 12 searches correction device

Claims (7)

(57) [Claims] 1. A word extracting means for extracting a word from an original sentence, a dictionary storing translation information necessary for translation of a headword of the original language, a part of speech classification, a translation of a target language, etc., and a word extracted by the word extracting means. Dictionary translation means for extracting translation information stored in the dictionary from the words, syntax analysis means for analyzing the syntax of the original sentence from the translation information obtained by the dictionary lookup means, and syntax analyzed by the syntax analysis means. A conversion / sentence generation unit for converting the translation information extracted by the dictionary lookup unit into a target language and generating a translated sentence, a word extracted by the word extraction unit, and a usage used in an original sentence .
And part of speech classification that, translation generated by the conversion-sentence generation means
Corresponds to the translation adopted in the sentence, and the part of speech class
A translated word correspondence list creating means for creating a translated word correspondence list by adding the frequency at which the translated word is used . 2. The machine translation apparatus according to claim 1,
The translated word correspondence list creating means may include a simple word translation list.
Addition of file information indicating the characteristics of the original text in which the word exists
A machine translation device for creating a word correspondence list . 3. A machine translation as claimed in claim 1 or claim 2.
The person who created the translated sentence in each translated word correspondence list created by the device
Merge their translation lists by adding their names,
Create an entire translation list by sorting words
A machine translation device comprising a list merging unit . 4. The machine according to claim 1, wherein
In the machine translation device, the word and the part of speech are the same but the translation is different.
Unified word search to search for words from translation list
Means, and from the translation generated by the conversion / sentence generation means,
Translation corresponding to the word searched by the unified word search means
Display means for detecting and sequentially displaying sentences, and
Translated words in the translation corresponding to the displayed words
A machine translation device , comprising: a correction unit that corrects a message . 5. The machine translation device according to claim 4, wherein
The display means is searched by the unified word search means.
Include sentences before and after the original and translated sentences that use the word.
A machine translation device characterized in that the message is displayed . 6. The machine translation device according to claim 5,
The display means, the original sentence and the words other single in translation
A machine translator characterized by displaying a word separately from words . 7. The machine according to claim 1, wherein
In the machine translation device, the word and the part of speech are the same but the translation is different.
Unified word search to search for words from translation list
Means and the translation of words searched by this unified word search means
Display means for displaying the word correspondence list, and a display on the display means.
The conversion corresponding to a specified word among the indicated words
・ Correct the translation of the translation generated by the sentence generation means
And a correcting means .