JPH06301714A - Machine translation system - Google Patents

Abstract

PURPOSE:To constitute the machine translation system so that a correct translated word can be selected without imposing a burden to a user. CONSTITUTION:An analyzing device 2 analyzes the original inputted through an input device 1, and delivers the result of analysis to a generating device 3 and a knowledge base controlling device 7. The generating device 3 translates the original based on a result of analysis by the analyzing device 2, the contents of a dictionary 5, knowledge information for selecting a translated word stored in a knowledge base 6, and a correcting instruction inputted from the input device 1 by a user, and outputs a translated sentence to an output device 4. The knowledge base controlling device 7 stores knowledge information for selecting a new translated word, or deletes knowledge information for selecting a translated word having a discrepancy from the knowledge base 6 based on the result of analysis by the analyzing device 2 and a translated sentence generated by the generating device.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a machine translation device, and more particularly to a machine translation device which can select a correct translated word without burdening the user.

[0002]

2. Description of the Related Art As a problem when performing machine translation
There may be multiple translations for a word. For example, the translation for Japanese "haku" is v
Omit, sweep, wear, etc. exist.

As described above, when there are a plurality of translated words for one word, it has been conventionally practiced to select a translated word according to the co-occurrence condition (for example, Japanese Patent Laid-Open No. HEI 2-
No. 301869). For example, for the word "haku", select "SWEEP" for "garbage removal" and "WEAR" for "wearing underwear". In this case, Fuku), a word related to the word, that is, a co-occurrence word that is a key of the co-occurrence condition (garbage, underwear in this example), and a translation (SWEE in this example).
(P, WEAR) is stored in the knowledge base for translation word selection, and the correct translation word is selected by using this translation word selection knowledge information.

[0004]

However, conventionally, since the user has to register the knowledge information for selecting the translated word in the knowledge base, there is a problem that the user is burdened.

An object of the present invention is to provide a machine translation device which can select a correct translated word without imposing a burden on the user.

[0006]

In order to achieve the above object, the present invention provides a knowledge base in which knowledge information for translation word selection is stored, a dictionary in which grammatical information and translation word information are stored, and contents of the dictionary. An analyzer for analyzing an original sentence, an analysis result of the analyzer, a generator for generating a translated sentence based on the contents of the dictionary, contents of the knowledge base, and a correction instruction, and managing the knowledge base. A knowledge base management device is provided, and the knowledge base management device changes the contents of the knowledge base based on the analysis result of the analysis device and the translated sentence generated by the generation device.

[0007]

When the original text to be translated is input, the analysis device analyzes the original text according to the contents of the dictionary, and the generation device analyzes the analysis device, the contents of the dictionary, the contents of the knowledge base, and the correction instruction. A translated sentence is generated based on and.

The knowledge base management device, for example, generates knowledge information for selecting a new translation word based on the analysis result of the analysis device and the translated sentence generated by the generation device, and the knowledge information for selecting the new translation word is the knowledge information. If it is inconsistent with the translation word selection knowledge information already stored in the base, delete the translation word selection knowledge information stored in the knowledge base that is inconsistent with the new translation word selection knowledge information. , If the new translated word selection knowledge information is not inconsistent with the translated word selection knowledge information already stored in the knowledge base, and the new translated word selection knowledge information is not stored in the knowledge base The contents of the knowledge base are changed based on the analysis result of the analysis device and the translated sentence generated by the generation device, such as storing new translation word knowledge information in the knowledge base.

[0009]

Embodiments of the present invention will now be described in detail with reference to the drawings.

FIG. 1 is a block diagram of an embodiment of the present invention. An input device 1, an analysis device 2, a generation device 3, an output device 4, a dictionary 5, a knowledge base 6 and a knowledge base management device. 7 and 7.

The input device 1 has a function of inputting an original sentence to be translated, a correction instruction and an end instruction.

The analysis device 2 has a function of analyzing the original sentence, the correction instruction and the end instruction input from the input device 1.

The generator 3 translates the original sentence analyzed by the analyzer 2 to generate a translated sentence, a function to correct the translated sentence according to a correction instruction, and a translated sentence to the knowledge base management device 7 according to an end instruction. It has a function of outputting.

The output device 4 has a function of outputting the translated sentence generated by the generating device 3.

The dictionary 5 stores grammatical information and translated word information necessary for analyzing an original sentence and generating a translated sentence.

The knowledge base 6 stores knowledge information for selecting translated words.

The knowledge base management device 7 has a function of managing the knowledge base 6.

Next, the operation of this embodiment will be described.

When translating, the user inputs the original sentence from the input device 1.

When the original sentence is input from the input device 1, the analysis device 2 grammatically analyzes the original sentence based on the grammatical information stored in the dictionary 5 to obtain the connection relation of each word.

Now, for example, if the original text input from the input device 1 is "I remove the garbage in the garden.", The analysis result of the analysis device 2 will be as shown in FIG. 2, for example.

The analysis result of the analysis device 2 is sent to the generation device 3 and the knowledge base management device 7.

When the analysis result of the original sentence is sent from the analysis device 2 to the generation device 3, the translation word of each word in the analysis result is converted into the dictionary 5.
It is obtained by referring to.

When there is one translated word for each word in the analysis result obtained by referring to the dictionary 5, the generating device 3 translates using the translated word obtained by referring to the dictionary 5.

If a plurality of translated words are obtained by referring to the dictionary 5 among the words in the analysis result, the generating device 3 outputs the translated word for each of the plurality of translated words. Create all combinations of words and one other word in the analysis results. After that, the generation device 3 passes the word for which a plurality of translated words is acquired and the combination including the word to the knowledge base management device 7, and acquires the translated word of the word for which the plurality of translated words is acquired from the knowledge base 6. To the knowledge base management device 7.

For example, as shown in FIG. 3, when a word having a plurality of translated words (foil, dust) is present in the analysis result, “fuku” and the combination “i, fuku”, "Trash, scrap" and "garden, scrap" are passed to the knowledge base management device 7 to request that the translated word "scrap" be taken out from the knowledge base 6, and "trash" and its combination ""I,trash","scrap,trash","garden,trash" are passed to the knowledge base management device 7, and it is requested that the translated word for "trash" be retrieved from the knowledge base 6.

When the above request is added from the generator 3,
The knowledge base management device 7 searches the knowledge base 6 for a combination included in the request. When the combination can be found, the translated word registered corresponding to the combination is notified to the generation device 3, and when the combination cannot be found, the generation device 3 is notified that the translated word is not registered.

When the knowledge base management device 7 notifies a translation word corresponding to a plurality of translation words stored in the dictionary 5, the generation device 3 determines the translation word notified from the knowledge base management device 7. It is used to perform translation, and the translated sentence is output to the output device 4.

When the knowledge base management device 7 notifies that the translated word is not registered, the generation device 3 translates using the translated word acquired by referring to the dictionary 5. At that time, for a word having a plurality of translated words, one of the translated words is selected and translated.

When the translated text is output to the output device 4, if the user needs to correct the translated text, the input device 1
Enter the correction instruction from, and if there is no need to correct it, enter the end instruction.

The correction instruction and the end instruction input from the input device 1 are sent to the generation device 3 via the analysis device 2.

When the correction instruction is input, the generator 3 corrects the translated sentence in accordance with the instruction and outputs the corrected translated sentence to the output device 4.

When the end instruction is input, the generation device 3 outputs the translated sentence to the knowledge base management device 7.

When the translated text is sent from the generation device 3, the knowledge base management device 7 decomposes the analysis result already sent from the analysis device 2 into a combination of two words,
Furthermore, the translated word corresponding to the combination is obtained from the translated sentence sent from the generation device 3.

After that, the knowledge base management device 7 stores the above combination and the corresponding translated word in the knowledge base 6.

After that, the knowledge base management device 7 searches the knowledge base 6 for the same combination as the combination stored this time.

Then, if it cannot be found, the processing is terminated.

If it is possible to find out, whether or not the translated word corresponding to the found combination is the same as the translated word corresponding to the combination registered this time, that is, the knowledge information for selecting the translated word stored this time and the translated word for finding the searched word are selected. Judge whether there is a contradiction in the knowledge information of.

If it is determined that they are the same, one of them is deleted from the knowledge base 6, and then the processing is terminated. Also,
If it is determined that they are different, both are deleted from the knowledge base 6, and then the process ends.

Now, for example, when the knowledge base 6 is in the initial state and no information is stored, the analysis result shown in FIG. 2 is input from the analysis device 2 and the translation sent from the generation device 2 is entered. Each word "I" in the analysis result based on the sentence,
The translations for "garden", "garbage", and "haku" are "I",
"GARDEN,""GARBEGE,""SWEE
Supposing that it is found to be "P", the knowledge base management device 7 first stores the contents shown in FIG. After that, the knowledge base management device 7 sets the knowledge base 6
And searched for the combination "I, Haku" registered this time,
It is determined whether or not "trash, scrap" and "garden, scrap" are already stored. In this case, since only the combination and the translated word stored this time are stored in the knowledge base 6,
The knowledge base management device 7 ends the processing. Therefore,
The contents of the knowledge base 6 are as shown in FIG.

After that, it is assumed that the user inputs the original sentence "I wear my underwear." From the input device 1.

When the original sentence "I wear my underwear." Is input, the analysis device 2 grammatically analyzes the original sentence based on the grammatical information stored in the dictionary 5 as described above, and then, as shown in FIG. The analysis result shown in (1) is passed to the generation device 3 and the knowledge base management device 7.

When the analysis result shown in FIG. 5 is sent from the analysis device 2, the generation device 3 acquires the translated word of each word in the analysis result by referring to the dictionary 5.

Now, for example, referring to the dictionary 5, it is assumed that the translated words shown in FIG. 6 have been obtained for each word in the analysis result.

The generator 3 uses a plurality of translated words "SWEEP",
Since the word "Haku" having "WEAR" and "VOMIT" exists in the analysis result, the word "Haku" for which a plurality of translated words are acquired and the combination "I, Pass "ku", "underwear, wear" to the knowledge base management unit 7,
Requests that the translated word for the word "haku" be retrieved from the knowledge base 6.

At this time, the contents of the knowledge base 6 are as shown in FIG. 4, and since the combination of "I, Haku" is stored in the knowledge base 6, the knowledge base management device 7 indicates "I, The translated word "SWEE" corresponding to the combination of "Haku"
The “P” is sent to the generation device 3 as a translated word corresponding to the word “haku”.

The generating device 3 obtains the translated word of the word "Haku" by referring to the dictionary 5 for the translated words of "SWEEP" passed from the knowledge base management device 7 and other words "I" and "underwear". Translation is performed as "I" and "UNDEWEAR", and the translated sentence is output to the output device 4.

However, in the case of "I wear my underwear."
Since the correct translation of “hak” is “WEAR”, the user inputs a correction instruction from the input device 1 to instruct to correct “SWEEP” into “WEAR”.
This correction instruction is sent to the generation device 3 via the analysis device 2.

The generating device 3 corrects the translated sentence according to the correction instruction, and outputs the corrected translated sentence to the output device 4.

When the user determines that the translated text output to the output device 4 is correct, the user inputs an end instruction.

This end instruction is sent to the generation device 3 via the analysis device 2, and when the generation device 3 receives the end instruction,
The corrected translated sentence (the translated sentence in which the translated word of “Haku” is corrected to “WEAR”) is passed to the knowledge base management device 7.

When the corrected translated sentence is sent from the generation device 3, the knowledge base management device 7 converts the analysis result shown in FIG. 5 already sent from the analysis device 2 into a combination of two words. It is decomposed, and a translated word corresponding to the combination is obtained based on the translated sentence sent from the generation device 3. After that, the knowledge base management device 7 stores the above combination and the corresponding translated word in the knowledge base 6.

In the case of this example, the knowledge base management device 7 uses "I, wear, I, WEAR", "underwear, wear, UNDE".
"RWEAR, WEAR", "I, underwear, I, UNDER
"WEAR" will be stored in the knowledge base 6. As a result, the contents of the knowledge base 6 are as shown in FIG.

After that, the knowledge base management device 7 searches the knowledge base 6 for the same combination as the combination "I, Haku", "Underwear, Haku", "I, Underwear" stored this time. In the case of this example, since the combination of "I, Haku" is searched, the knowledge base management device 7 corresponds to the translated word "I, Haku" and the combination "I, Haku" stored this time. It is determined whether or not the translated word corresponding to is identical, that is, whether or not there is a contradiction between the two pieces of knowledge information.

In the case of this example, the translated word corresponding to the combination "I, Haku" stored this time is "I, WEAR", and the translated word corresponding to the searched combination "I, Haku" is "I, SWEEP". Therefore, it is determined that the two pieces of knowledge information are inconsistent, and the two pieces of knowledge information are added to the knowledge base 6
Remove from.

As a result, the contents of the knowledge base 6 are as shown in FIG.

By repeating such processing, only knowledge information for translation word selection including a co-occurrence word which is a key for translation word selection remains in the knowledge base 6.

[0058]

As described above, according to the present invention, the knowledge base for changing the contents of the knowledge base in which the knowledge information for translation word selection is stored based on the analysis result of the analysis device and the translated sentence generated by the generation device. Since the management device is provided and the knowledge information for selecting the translated word can be automatically stored in the knowledge base, there is an effect that the correct translated word can be selected without burdening the user.

[Brief description of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention.

FIG. 2 is a diagram showing an example of an analysis result of the analysis device 2.

FIG. 3 is a diagram showing an example of translated words of respective words existing in a dictionary 5.

FIG. 4 is a diagram showing an example of contents of a knowledge base 6.

5 is a diagram showing an example of an analysis result of the analysis device 2. FIG.

FIG. 6 is a diagram showing an example of translated words of respective words existing in the dictionary 5.

7 is a diagram showing an example of contents of a knowledge base 6. FIG.

FIG. 8 is a diagram showing an example of contents of a knowledge base 6.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 ... Input device 2 ... Analysis device 3 ... Generation device 4 ... Output device 5 ... Dictionary 6 ... Knowledge base 7 ... Knowledge base management device

Claims (3)

[Claims] 1. A knowledge base in which knowledge information for translation word selection is stored, a dictionary in which grammar information and translation word information are stored, an analysis device for analyzing an original sentence according to the contents of the dictionary, and an analysis result of the analysis device. A knowledge base management device that manages the knowledge base, and a generation device that generates a translated sentence based on the contents of the dictionary, the contents of the knowledge base, and a correction instruction. A machine translation device, characterized in that the contents of the knowledge base are changed based on the analysis result of the analysis device and the translated sentence generated by the generation device. 2. The knowledge base management device generates knowledge information for new translation word selection based on the analysis result of the analysis device and the translated sentence generated by the generation device, and selects the new translation word selection information. If the knowledge information of the word is inconsistent with the knowledge information for selecting a translation word stored in the knowledge base, it is inconsistent with the knowledge information for selecting a new translation word For selecting the translation word stored in the knowledge base Of the new translation word selection information is deleted, the new translation word selection knowledge information is consistent with the knowledge information stored in the knowledge base, and the new translation word selection knowledge information is stored in the knowledge base. The machine translation device according to claim 1, wherein the knowledge information for selecting the new translated word is stored in the knowledge base if not. 3. The translation word selection knowledge information includes a translation target word, a co-occurrence word for the translation target word, and a translation word for the translation target word. Item 2. The machine translation device according to item 2.