JPH06203059A - Machine translation apparatus - Google Patents

Abstract

PURPOSE:To heighten translation accuracy by performing equivalent selection/ learning in batch fashion without using a post-edit function by interactive translation of one text unit, acquiring the part of speech information of a word in an original text by the equivalent selection, and utilizing it in translation processing. CONSTITUTION:This apparatus is provided with a display means 6 which displays the word in a translation target document with corresponding equivalents>=one, an equivalent instruction means 7 which selects one desired equivalent from displayed equivalents, an equivalent information learning means 9 which stores learning information with respect to an instructed equivalent, and a translation execution means 11 which performs translation by using the instructed equivalent in addition to a conventional English-Japanese translation apparatus. Also, it is provided with a part of speech information estimation means 8 which estimates the part of speech of the word from the equivalent instructed by a user, and the original text can be analyzed and translated based on estimated part of speech information.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translator.

[0002]

2. Description of the Related Art In recent years, machine translators have been widely used as devices for translating one language into another. In addition, positioning as a translation support device is becoming standardized without being particular about automatic translation. However, the current technical level of machine translation is limited, and in very long sentences and sentences that cannot be deciphered without considering the meaning (for example, novels), the quality of the translation result is significantly reduced, Not enough to use as is. In particular, it is very difficult to automatically determine the translated word with high accuracy (translated word determination processing), and it is not expected that this problem will be solved in the near future. Then, in order to solve the problem of determining the translation word, the conventional translation provides the user with the following two methods (functions). One is "user dictionary registration function"
The other is the "learning function" through the "translated word selection function".

The "user dictionary registration function" is a function that allows an arbitrary translated word to be set for an arbitrary word specified by the user, and words with the same heading are the basic dictionary, the technical term dictionary, and the user dictionary. When registered in each of the above, the translation registered in the user dictionary is preferentially adopted (translated). When translating a certain document, the user registers the translation of each word in the user dictionary in advance before translating it, and corrects the word of the same heading to the same translation many times using an editor. It is released from the trouble of doing.

Next, the "translation word selection function" refers to a word that composes a translated sentence (translation result) translated by the translator, presents another translation word of the word designated by the user, and is suitable for the user. This is a function to select a translation word (translation word selection function). The new translation selected in the translation will replace the original translation. In addition, the “learning function” means remembering the translated word information for a word for which the user has selected a translated word through the translated word selection function, and then, for a certain predetermined period after that, the translated word for that word It is a function that is automatically applied and translated. This function also works when translating a document in which the same word appears many times, because it is necessary to make that word remember the desired translated word once in the translated word selection operation. This frees you from the trouble of using the editor to modify the same translation into different translations many times.

Next, regarding translation processing, particularly in English-Japanese translation, in order to correctly analyze a sentence, it is necessary to clarify in advance what part of speech each word exists in the sentence. There is. Since English is a multi-part-of-speech word, this problem is considered as the "multi-part-of-speech elimination problem", which is one of the problems that is very difficult to solve in English-Japanese translation. There is also the idea of comprehensively preparing possible combinations of parts of speech, analyzing and translating all of them, and obtaining the result, but the number of combinations easily explodes, and unnecessary candidate branching technology is required. . In any case, this problem of multi-part-of-speech elimination is a difficult problem, and if even a few words in the original sentence have a clear part-of-speech interpretation in advance, there is no doubt that the translation accuracy will increase.

[0006]

By the way, it is not possible for the user to know whether the translated word for a certain word is appropriate or inappropriate unless it is translated once. Therefore, it takes a considerable amount of time to translate a large amount of text. Also,
If an incorrect translation is obtained, either because the desired translation does not exist in any dictionary, or because it has a low priority even if it exists, unless the translation information for that word is displayed. Not clear.

Generally, a plurality of user dictionaries can be used as a translation environment, but even in that case, it is only possible to set the priority order in dictionary units. For example, when there are a user dictionary 1 and a user dictionary 2 and word A and word B are registered in both user dictionaries, word A is a translation of user dictionary 1 and word B is a translation of user dictionary 2. Even if you think about it, you cannot set such a priority because you can set priorities only in dictionary units. In such a case, it is best to set a translated word for each of the words A and B by the learning function. However, in order to select a translated word, interactive translation is performed on a sentence-by-sentence basis, and on the assumption that translated word selection is used as a function of post-editing, the conventional translator has no method. One-sentence-based translation basically requires the user to paste it into the translator under the assumption that all sentences are translated, and is very redundant and is not used for users who mainly perform batch-like translation without time. There are many cases. Further, since the translated word learning function acquires learning information only through the translated word selection operation, a user who does not perform the translated word selection operation in the one-sentence interactive translation cannot benefit from the learning function.

Next, regarding the problem of part-of-speech interpretation of each word in translation, in most cases, the part-of-speech interpretation in the original sentence and its part-of-speech in the translated sentence are the same. For example, the English word plan translates as a noun: Plan 2. Plan 3. Method 4. There are drawings, etc., and as verbs, 1. Plan 2. There is a plan, but the original text is It is a good plan to go at o
nce. "If it's a good plan to go right away," that is, p
If lan is a noun, the translation "plan" is a noun even in the translated sentence, and the original sentence is I plan a new book on chemistr.
y. If “plan a new book on chemistry”, that is, if plan is a verb, then in the translation, the translation “plan” is a verb. Nevertheless, in the conventional translator, since the user selects the translated word after the translation processing in the interactive translation, the part-of-speech information obtained there cannot be used effectively in the translation processing.

The present invention solves such a conventional problem, and realizes batch word selection / learning without using a post-editing function by interactive translation of one sentence unit, and It is an object of the present invention to provide a machine translator in which part-of-speech information of a word in an original sentence is acquired through translation word selection and used in translation processing to improve translation accuracy.

[0010]

In order to achieve the above object, the present invention searches a word used in a translation target sentence based on a translation dictionary, and based on the result of the search A list of words used in is displayed with one or more translated words corresponding to each word, and the user can select an appropriate translated word from the displayed multiple translated words, and the specified translated word information. Is stored together with the corresponding word information, and in the subsequent translation, the translated word already stored for that word is used for translation.

Before the actual translation of the sentence to be translated, translation word selection processing can be performed, and the part of speech of the word in the original sentence is estimated from the translation word information obtained there and used for translation (analysis) processing. It was done like this.

[0012]

With the above-described structure, the present invention allows the user to collectively perform the operation related to the translation word selection without performing the interactive translation of each sentence. Further, due to the effect of learning, when the same word appears after that, the desired translated word automatically appears, so that the labor of the translation work can be reduced.

Further, through the translation word selection processing, since the a priori information about the part-of-speech interpretation processing of each word is obtained, the analysis ability of the original sentence is improved, and as a result, the translation accuracy (translation quality) itself can be expected to be improved.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. The present invention is not necessarily intended for an English-Japanese machine translator, but in the following description, an English-Japanese machine translator is premised for concrete explanation.

FIG. 1 is a functional block diagram of a machine translator according to an embodiment of the present invention. 1 is the document to be translated (original)
Input means such as a keyboard, floppy disk, network, etc. for inputting, 2 is an original / translated text storage means for storing the original text input from the input means 1 in association with the translation result, and 3 is a vocabulary necessary for translation It is a dictionary for translation such as a basic dictionary, a technical term dictionary, and a user dictionary that provides knowledge about. By the way, the basic dictionary is always searched in normal translation, and there are actually multiple technical term dictionaries such as information processing term dictionary, scientific term dictionary, machine term dictionary, etc. Is specified. The user dictionary is a dictionary that the user can create,
There are a plurality of them, and like the technical term dictionary, the user arbitrarily specifies them at the time of translation. 4 is based on the translation term dictionary 3,
It is a dictionary search means for searching words (including compound words) in the translation target document stored in the original / translated text storage means 2. Reference numeral 5 is a word information storage means for storing the word information retrieved by the dictionary retrieval means 4. However, at that time,
If the word is already registered, no new registration is performed to avoid duplication. The stored word information includes a translated word of the searched word and texts before and after the sentence including the searched word. Reference numeral 6 is an original sentence or a translated sentence stored in the original sentence / translated sentence storage means 2 or a word retrieved by the dictionary retrieval means 4 and stored in the word information storage means 5 with associated information (translated word, text before and after each word, etc.). ) And a display means for displaying. 7
Is a translation word indicating means such as a keyboard or a mouse for selecting a translation word desired by the user with respect to a translation word of each word displayed on the display means 6, and 8 is associated with the translation word information specified by the user by the translation word indicating means 7. It is a part-of-speech information estimation means for estimating the part-of-speech information of a word. Reference numeral 9 is a translated word information learning unit that stores (translated word) learning information regarding a word designated by the user by the translated word designation unit 7 and the estimated part-of-speech of the word estimated by the part-of-speech information estimation unit 8. Learning information means, for example, "teacher" translates as "teacher",
This is information about translated words. Reference numeral 10 is a translation rule that provides knowledge about the grammar necessary for performing translation. Reference numeral 11 is a translation execution means for translating a given document to be translated while referring to the knowledge of the translation dictionary 3 and the translation rule 10. The translated text obtained by translation is the original
It is stored in the translated text storage means 2. Further, depending on the form of the translation function, it is displayed on the display unit 6. 12 is a preposition or b
It is an independent word filtering means for removing non-independent vocabulary subordinate to other parts of speech such as e verb.

FIG. 2 is a circuit block diagram of this embodiment.
Reference numeral 21 is a keyboard (including a mouse), 22 is a central processing unit (hereinafter abbreviated as CPU), 23 is a cathode ray tube display (hereinafter abbreviated as CRT), 24 is a random access memory (hereinafter abbreviated as RAM). 25)
Is a read-only memory (hereinafter abbreviated as ROM).

The input means 1 and the translation word indicating means 7 shown in FIG.
Using the keyboard 21, the original / translated sentence storage means 2, the translation rule 10, the word information storage means 5, the translated word information learning means 9
By the RAM 24, the display means 6 by the CRT 23,
In the translation executing means 11, the CPU 22 has the RAM 24 and the RO.
It is realized by executing the program stored in the ROM 25 while exchanging data with the M25. The translation dictionary 3 includes a ROM 15 and a RAM 2
4 or a secondary storage device.
The dictionary search means 4 and the part-of-speech information estimation means 8 are realized by the CPU 22 executing the programs stored in the ROM 25.

FIG. 3 is a flow chart showing the flow of processing in one embodiment of the present invention. The flow of processing in the embodiment of the present invention will be described below with reference to this flowchart. First, in step 31, the input means 1 (keyboard, floppy disk, network, etc.)
The original text (document to be translated) is input. The input original sentence is stored in the original sentence / translated sentence storage unit 2.

Next, in step 32, the user activates the batch-type translated word selection function, which is the substance of the present invention, through the input means 1.

Next, in step 33, the words in the original sentence stored in the original sentence / translated sentence storage means 2 are retrieved by the dictionary retrieval means 4. From the retrieved words, the independent word filtering unit 12 excludes the vocabulary of the part of speech that is basically difficult to select a translated word, such as prepositions and be verbs, and stores only the independent words in the word information storage unit 5. . For example, for the sentence I wanted to be a teacher.
When the search / registration process is executed, I, want, to, be, a, tea
cher is searched as a word, but "to, be, a" is excluded by "independent word filtering means 12", and I, want, te
Word information about acher is stored. Further, when newly registering word information, it is checked whether or not information on the word has already been registered in the word information storage means 5 to avoid duplication of registration for words of the same heading.
Therefore, after the search and registration process for the sentence I wanted to be a teacher.
When the sentence r, too. is subject to the search / registration process, he, want, to, be, a, teacher, too are newly searched as words, but by the "independent word filtering means 12", First, to, be, a, too are excluded and left, he, wan
For t and teacher, since want and teacher have already been registered, as a result, only the word information regarding he is stored.

Further, in this case, when the dictionary search means 4 searches for a word in a document, the text including the searched word and showing the context before and after the word, for example, KWIC, is simultaneously extracted to store the word information. The means 5 stores the text together with the corresponding word, and the display means 6 displays the text together with the searched word, so that the context of the word designating the translated word is implied, so that the user can easily select the appropriate word. It is possible to specify a translation. That is, the document to be searched is not changed until the batch-type translated word selection function in this embodiment is completed, so that the text information is stored depending on the text position in the document. For example, the two sentences above, I want
ed to be teachwer.He wanted to be a teacher, too.
However, suppose it is at the beginning of the sentence, and when memorizing the word information about He in the second sentence, ed to be a teacher.
If a text such as “He wanted to be a teac” is stored and can be displayed to the user, the user can easily determine the translated word regarding He. In this case, 20 characters before and after He are displayed.
It is enough to remember the text position of.

Next, at step 34, the word information storage means 5
The word information stored in is displayed by the display means 6 as shown in FIG. In this display, the default translation is shaded. In the case of FIG. 4, teach
The default translation of r is "teacher". that time,
When the number of actually searched words is larger than the display area of the translator, the user refers to all the word information through a user interface such as a scroll bar or a page break function. Further, as described above, the text information in which each word exists may be displayed at the same time.

Next, in step 35, the translation word instructing means 7 indicates the translation word desired by the user from the translation word information of the word displayed by the display means 6. Figure 5 shows tea
The result of selecting “teacher” as the translation of cher is shown. The highlighting changes from "teacher" to "teacher." The translated word information thus designated is stored by the translated word information learning means 9. In the selection example shown in FIG. 5, the translation of “teacher” is “teacher”. The translated word information is stored in the translated word information learning means 9.

Further, in this embodiment, the part-of-speech information estimating unit 8 estimates the part-of-speech of the original word from the translated word designated by the user. If the part of speech of the translated word selected by the user is a noun, the part of speech of the original word is also a noun, and if the part of speech of the translated word selected by the user is a verb, the translated word of the original word is estimated to be a verb. The result of the selection in FIG.
The part of speech of "er" is a noun. The estimated part-of-speech information is stored in the translated word information amount sort means 9. Through this translated word selection interface, the user can select translated words at once for any displayed word. When the translation selection by the user is completed, the batch translation selection function ends.

Next, in step 36, the user instructs the translation. The translation is executed by the translation executing means 11. In the above example, it is originally translated as "I wanted to be a teacher. He also wanted to be a teacher." However, based on the translated word information stored by the translated word information learning means 9, "I wanted to be a teacher.
He also wanted to be a teacher. A translated sentence is generated. Furthermore, by using the part-of-speech information stored in the translated word information learning means 9 to analyze and translate the original sentence, the ability to analyze the original sentence is improved, and as a result, the translation accuracy itself can be expected to be improved.

As described above, according to the above-described embodiment, by providing a function for implementing translation word selection in batch and a user interface, it is possible to provide translation word selection and learning functions for users who do not like interactive translation. It becomes available. In addition, translation accuracy can be improved by estimating the part of speech of a word through translation selection and using it for analysis of the original sentence.

[0027]

As is apparent from the above-described embodiment, the present invention can select a translated word without using the post-editing function in the interactive translation of one sentence unit, and improve the operability for batch translation users. it can. In addition, since the part-of-speech information is obtained for the word for which the translated word is selected, the analysis accuracy of the original sentence is improved, and as a result, the translation quality can be expected to be improved.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a machine translator according to an embodiment of the present invention.

FIG. 2 is a circuit block diagram of a machine translator in the embodiment.

FIG. 3 is a flowchart showing an operation of the machine translator in the embodiment.

FIG. 4 is a list view showing a display example when selecting a translated word in the embodiment.

FIG. 5 is a list diagram showing a display example when a translation other than the default translation is selected when selecting a translation word in the embodiment.

[Explanation of symbols]

 1 input means 2 original sentence / translated sentence storage means 3 translation dictionary 4 dictionary retrieval means 5 word information storage means 6 display means 7 translated word instruction means 8 part-of-speech information estimation means 9 translated word information learning means 10 translation rules 11 translation execution means 12 independent word filtering Means 21 Keyboard 22 CPU 23 CRT 24 RAM 25 ROM

Claims (3)

[Claims] 1. An original sentence / translated sentence storage unit for storing a translation target document and a translation result in association with each other, a translation dictionary for storing a translation of a word, and a word used in a translation target sentence as a translation dictionary. A dictionary search means for performing a search based on the word information, a word information storage means for storing the searched word information, and a list of words used in the document based on the contents of the word information storage means, corresponding to each word. Display means for displaying together with more than one translation word, translation word instructing means for instructing a suitable translation word from the one or more translation words displayed by the display means by the user, and learning information about the translation word instructed by the user A machine translation machine for translating a document to be translated while preferentially using learning information about the translated word for translation of words. . 2. The machine translator according to claim 1, wherein when the dictionary search means searches for a word in a document, a text including the searched word and extracting contexts before and after the word are simultaneously extracted. A machine translator, wherein the word information storage means stores the text together with the corresponding word, and the display means displays the text together with the retrieved word. 3. The machine translator according to claim 1, further comprising a part-of-speech information estimation unit that estimates the part-of-speech of the original word from the user's instruction information regarding the translation word selection obtained by the translation word instruction unit, and the translation execution unit. A machine translator in which the part-of-speech information estimated by the part-of-speech information estimating means is utilized for analysis of an original sentence and translated.