JPH0997256A - Dictionary registration device and machine translation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To enable the automatic preparation and the automatic setting of a document corresponding dictionary by the unit according to the feature of a document by extracting the feature of an inputted document, deciding the preparation necessity of the document corresponding dictionary, preparing the document corresponding dictionary and perform a registration operation. SOLUTION: A document feature extraction part 9 extracts document feature information from an inputted Japanese document. A document corresponding dictionary deciding part 11 calculates the degree of similarity of the extracted document feature information and the dictionary feature information corresponding to each document corresponding dictionary 16, 17,... within a dictionary feature data base 10 and decides the document corresponding dictionary of the dictionary feature information having the maximum degree of similarity from the dictionary feature data base 10. When plural document corresponding dictionarys of the maximum degree of similarity exist, they are allowed to be selected by a user. When the document corresponding dictionary to be registration object is determined by including the case where the document corresponding dictionary is prepared in this way, the dictionary name determined by the document corresponding dictionary decision part 11 is set to the dictionary registration part 14 of a dictionary interface part 4. When the user inputs a word to be registered, a word registration is performed for the document corresponding dictionary having this dictionary name.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dictionary creating device and a machine translation device, and more particularly to a device having a function of creating a plurality of individual dictionaries and using them.

[0002]

2. Description of the Related Art In a machine translation apparatus, in order to maintain high translation quality (quality of translated text), it is essential to use a dictionary or a technical term dictionary created individually by a user. Various devices having a dictionary have been proposed. Even if the machine translation device holds multiple individual dictionaries and technical term dictionaries, in order to achieve higher translation quality, the most appropriate dictionary for the input document is selected from among multiple types of dictionaries. It is necessary to choose and use that dictionary.

Conventionally, the user of the machine translation device has selected the dictionary by himself / herself. However, if the machine translation device has a function of automatically selecting an optimal dictionary for an input document, the translation work can be made efficient and high-quality translation results can be obtained.

Reference 1 “Japanese Patent Laid-Open No. 6-332946” In the machine translation apparatus described in Reference 1, as a method of automatically selecting an optimal dictionary for a document to be translated from a plurality of types of dictionaries, input is performed. A method has been proposed in which a context vector described in a basic word dictionary is extracted from a document and the specialized field of the input document is determined from the context vector. That is, each dictionary is assigned a code of the specialized field of the context vector related to the dictionary, and a method of selecting a dictionary having the same specialized field code as the specialized field (context vector) determined from the input document is used. I am taking it.

[0005]

However, according to the dictionary selection method described in Document 1, since the field code of the specialized dictionary is fixed when the dictionary is created, only the dictionary can be selected within the fixed specialized field.

That is, only the field codes (for example, "chemistry" and "information processing") of technical terms given to the dictionary at the time of creating the individual dictionary can be estimated from the context vector obtained from the input document. It is not possible to classify or select a dictionary according to the characteristics of the input document other than the field code. For example,
Considering the translation of the word "structure", even in the field of "information processing", depending on the creator and content of the document, "structur"
It may be translated as "e" or as "construction", and these cannot be translated by the classification of "information processing".

Generally, when a user creates a dictionary, a document to be translated exists, and a dictionary necessary for translation processing is created in order to improve the translation quality of the document or a document similar to the document. . Therefore, it is desirable that the dictionary is not prepared in a fixed classification such as a specialized field, but is prepared and selected in a classification based on the type and characteristics of the document to be translated.

Therefore, there is a demand for a dictionary registration device capable of creating and registering a dictionary so that an optimum dictionary can be selected for a document to be translated without adding a specialized field code, and such a dictionary is effectively used for translation. There is a need for a machine translation device that can be used.

[0009]

In order to solve such a problem, in the first aspect of the present invention, a dictionary registration device for creating and registering a dictionary used in a machine translation device has the following means. Is characterized by.

That is, the dictionary registration device according to the first aspect of the present invention extracts (A) document feature information including at least appearance frequency information of words and idioms in an input document described in natural language. Document feature extraction means,
(B) For each of the one or more document-corresponding dictionaries that have already been created, a dictionary feature storage unit that stores dictionary feature information having a format similar to the document feature information, and (C) a document feature extraction unit. Obtain the similarity between the document feature information created and each dictionary feature information stored in the dictionary feature storage means,
Based on the obtained similarities, it is necessary to create a new document correspondence dictionary and, if there is no need to create a new document correspondence dictionary, determine an existing document correspondence dictionary to be registered. , To create a new document correspondence dictionary,
The extracted document feature information is stored in the dictionary feature storage means as the dictionary feature information, and when it is determined that the existing document corresponding dictionary is to be registered, the dictionary is stored according to the extracted document feature information. Similarity determination means for updating the dictionary feature information stored in the feature storage means, and (D) based on the determination result of this similarity determination means, if necessary, after creating a new document correspondence dictionary, the determination is made. And a dictionary registration means for registering a dictionary in the document corresponding dictionary.

According to the dictionary registration device of the first aspect of the present invention,
It is possible to automatically create and automatically set the document correspondence dictionary in units according to the characteristics of the input document, not in fixed field units.

In a second aspect of the present invention, a machine translation apparatus having a dictionary registration structure is provided with the above-mentioned means (A) to (D) and further has the following means.

That is, the machine translation apparatus of the second aspect of the present invention is (E) a second feature of extracting at least document feature information from the translation target input document, the appearance frequency information of words and idioms in the translation target input document. The similarity between the document feature extraction means, (F) the document feature information extracted by the second document feature extraction means, and each dictionary feature information stored in the dictionary feature storage means is obtained, and each obtained similarity is obtained. Second similarity determining means for determining one or more document correspondence dictionaries to be used in the translation process based on the degree, and (G) the translation target input document to the document correspondence dictionary decided to be used in the translation process. It further has a translation executing means for translating the contents.

Since the machine translation apparatus of the second aspect of the present invention extracts the document feature from the document to be translated and selects the optimum one from a plurality of document correspondence dictionaries in units according to the feature of the document. Even if the same word is used, the translation can be performed according to the input translation target document, and the translation quality can be further improved as compared with the conventional device that prepares a dictionary in fixed field units.

[0015]

BEST MODE FOR CARRYING OUT THE INVENTION An embodiment in which the present invention is applied to a Japanese-English machine translation device will be described in detail below with reference to the drawings.

The Japanese-English machine translation apparatus of this embodiment has a dictionary creating and registering function, and its implementation structure is as follows:
It constitutes one embodiment of the dictionary creating apparatus according to the present invention. That is, the machine translation device of this embodiment has a dictionary registration mode and a translation mode as operation modes.

(A) First Embodiment The machine translation apparatus of the first embodiment is actually an input device such as a keyboard and a mouse, an output device such as a CRT display, a liquid crystal display and a printer, a hard disk device and the like. It is composed of a workstation provided with an auxiliary storage device, an information processing device such as a minicomputer, a personal computer, etc. However, from the feature of this embodiment, when the main part configuration is divided into functional parts, a functional block diagram shown in FIG. 1 is obtained. It has the structure shown.

In FIG. 1, the machine translation apparatus of the first embodiment is a user interface section 1 for the user to input a document to be translated and to display the translation result to the user, and an input document for dictionary creation. Or a feature of a translation target document is extracted to determine whether or not to create a dictionary and to determine a dictionary to be used, a dictionary determination unit 2, a request such as registration to the dictionary determined by the dictionary determination unit 2, or a determination by the dictionary determination unit 2. A dictionary interface unit 4 for extracting dictionary contents used for translation processing from the created dictionary, a translation execution unit 3 for executing translation processing, and
The dictionary storage unit 5 stores a plurality of dictionaries.

The user interface section 1 is a dictionary creating section 6 that serves as a user interface when a user registers and creates a dictionary, a document input section 7 that receives a document to be created or translated, and , And a translation result output unit 8 for presenting and outputting the translation result to the user. The information on the dictionary registration mode or the translation mode is, for example, taken in before the document input unit 7 takes in the input document. Further, in the case of the first embodiment, the document input unit 7 is adapted to take in only the document of the original language Japanese in the dictionary registration mode, and in the translation mode, It is designed to import Japanese documents to be translated.

The dictionary determination unit 2 is composed of a document feature extraction unit 9, a dictionary feature database (dictionary feature DB) 10, and a document correspondence dictionary determination unit 11.

In the dictionary registration mode, the document feature extraction unit 9 extracts the document features from the Japanese document relating to the source language. In the translation mode, the document feature extraction unit 9 extracts the document feature information from the Japanese document to be translated.

The document feature extraction unit 9 executes morphological analysis and the like in order to extract the document feature information. Although dictionary contents are required for this analysis and the like, for example, at least the contents stored in a general-purpose dictionary 19 described later are used. Morphological analysis or the like may be used for all the stored contents stored in the dictionary storage unit 5. Further, the document feature extraction unit 9 has a built-in dictionary unit that stores the dictionary contents transferred from the dictionary storage unit 5 necessary for such processing so as to be directly accessible for such processing. Is also good.

The dictionary feature database 10 stores feature information (hereinafter, referred to as dictionary feature information) for each document corresponding dictionary stored in the dictionary storage unit 5.

In both the dictionary registration mode and the translation mode, the document-corresponding dictionary determining unit 11 is most similar to the document feature information extracted by the document feature extracting unit 9 from the dictionary feature information in the dictionary feature database 10. The existing document correspondence dictionary, which will be described later and has the dictionary feature information, is determined. In the dictionary registration mode, the document-corresponding dictionary determining unit 11 further determines the necessity of creating the document-corresponding dictionary related to the current input document according to the determination result. In the dictionary registration mode, the document-corresponding dictionary determination unit 11 appropriately executes the update operation for the dictionary feature database 10 based on the similarity determination result and the like.

FIG. 2 is an explanatory diagram of an example of the document characteristic information. The document feature information 202 of this example is basically the input Japanese document as shown in FIG. 2A as shown in FIG. 2B regardless of the dictionary registration mode or the translation mode. This is a frequency distribution 2024 of words of a predetermined part of speech (for example, a noun) that appears a predetermined number of times or more (for example, five times or more) in the source language document 201. In this example, the file name (filename) 2021, the editor information (editor) 2022, and the user information (user) 2023 input along with the input document are input.
Etc. and the bibliographic information also constitute the document feature information.

FIG. 3 is an explanatory diagram of an example of dictionary feature information stored in the dictionary feature database 10. FIG.
Are three types of document formation dictionaries 1 in the dictionary storage unit 5 to be described later.
Information examples 301, 302, 30 for 6, 17, 18
3 is shown, and its configuration is similar to the document feature information, as is clear from the comparison with FIG.

The document-corresponding dictionary determination unit 11 described above determines, for example, that the similarity between the document feature information and the dictionary feature information is the number of words (nouns) common to both feature information, and the word belonging to the document feature information is the most. It is determined that the document forming dictionary containing many documents is the most similar.

The translation execution unit 3 functions only in the translation mode, and is composed of a translation processing unit 12 for executing translation processing and a dictionary unit 13 for storing dictionary contents used for translation processing.

The translation processing unit 12 is the same as that installed in the existing machine translation device, and therefore its detailed description is omitted. Note that the translation processing unit 12 has a morphological analysis unit, a syntactic analysis unit, and the like for the source language document (Japanese document) in a more detailed view. It can be used in common with the department and the parsing unit. The dictionary unit 13 is for transferring a predetermined dictionary content, which will be described later, from the dictionary storage unit 5 and storing it.

The dictionary interface unit 4 comprises a dictionary registration unit 14 and a translation interface unit 15.

The dictionary registration unit 14 functions in the dictionary registration mode, and based on the judgment result of the document registration dictionary judgment unit 11 for the document for dictionary registration, the contents are additionally registered in the existing document correspondence dictionary, The document corresponding dictionary is newly created and the contents are registered in the newly created document corresponding dictionary.

The translation interface unit 15 functions in the translation mode, and transfers the stored contents of the dictionary determined by the document correspondence dictionary determination unit 11 for the translation document to the dictionary unit 13 in the translation execution unit 3. It is what makes me.

The dictionary storage unit 5 stores a plurality of document-corresponding dictionaries 16, 17, 18, ... With different features, and a general-purpose dictionary 19. The document correspondence dictionaries 16, 17, 18,
... stores the dictionary contents formed from a dictionary registration document which is a Japanese document and an English document. On the other hand, the general-purpose dictionary 19 stores general-purpose dictionary contents in many translation target documents.

FIG. 4 is a flow chart showing the operation in the dictionary registration mode in the machine translation device of the first embodiment.

When the processing of the dictionary registration mode is entered, the document input section 7 takes in the Japanese document which is the object of the dictionary creation inputted by the user (step 402). Then, the document feature extraction unit 9 extracts the document feature information (20) from the input Japanese document.
2) is extracted (step 403).

The document correspondence dictionary determination unit 11 extracts the extracted document feature information (202) and the dictionary feature information (301, 302, corresponding to each document correspondence dictionary 16, 17, 18, ... In the dictionary feature database 10). (303, ...), the document correspondence dictionary of the dictionary feature information having the maximum similarity is determined from the dictionary feature database 10 (step 404).

Here, in order to simplify the explanation, it is assumed that the method of calculating the similarity of document characteristic information is as follows. Note that, as described above, the document feature information and the dictionary feature information have a set of words (frequency distribution) appearing five times or more as the central information.

First, when there is dictionary feature information having the same file name as the file name in the document feature information, the degree of similarity is infinite. If there is no dictionary feature information having the same file name, words described in both the document feature information and the dictionary feature information (words that appear 5 times or more)
The number of is the similarity. However, if the number of words described in both the document characteristic information and the dictionary characteristic information is less than or equal to a predetermined number (may be 0), the degree of similarity is 0.

In the range of information shown in FIGS. 2 and 3, the dictionary feature information 301 has a similarity of 2 to the document feature information 202 (“information” and “technology” are duplicated),
The dictionary feature information 302 has a similarity of 3 to the document feature information 202 (“information”, “research”, and “technology” overlap), and the dictionary feature information 303 has a similarity of 1 to the document feature information 202 (“technology”). Is the dictionary feature information 302), and the document corresponding dictionary 17 having the dictionary feature information 302 is determined as the dictionary having the maximum similarity.

If there is no document correspondence dictionary having dictionary feature information similar to the document feature information related to the input document, a new document correspondence dictionary is created in the dictionary storage unit 5 (step 406). At this stage, only the framework (file) of the dictionary is formed, and the contents are registered later as described later.

On the other hand, when there are a plurality of document correspondence dictionaries having the highest degree of similarity, one is not automatically selected, but
All the candidate dictionaries are presented to the user through the dictionary creating unit 6 and the user is made to select them (step 407).

When the document-corresponding dictionary to be registered is determined, including the case of being created as described above, the dictionary name determined by the document-corresponding dictionary determining unit 11 is used as the dictionary registration unit of the dictionary interface unit 4. When it is set to 14, and the user inputs a word to be registered (step 408), the word is registered in the document corresponding dictionary having this dictionary name (step 409).

Any specific registration method may be used. For example, it is determined whether or not a word included in the document feature information that appears five times or more is described in the document-corresponding dictionary to be registered. If not, the word is presented to the user. Confirm whether or not to register, and when registering, import the English information and register. In this case, the number of appearances less than 5 may be targeted. Also, the words may be registered in the dictionary regardless of the feature extraction. Therefore, the registered word may be other than the word of the part of speech related to the feature extraction. For example, the contents that can be stored in the dictionary from the bilingual document may be obtained in advance by the method described in Document 2 and registered by such a registration operation.

Reference 2 “Mihoko Kitamura and Yuji Matsumoto,“ Automatic acquisition of translation knowledge from bilingual bilingual corpus ”, Report of NLC workshop of IEICE, Technical Report Vol.94 No.32
(2), pp.9-16 ”Here, the document correspondence dictionary 17 is the dictionary interface unit 4
When it is set in the dictionary registration unit 14 of the user, the user selects the “structure: structure” for the “structure” in the document 201.
Is registered in the document correspondence dictionary 17.

When the user gives an instruction to end the registration, the dictionary feature database 10 is updated (step 410), and the processing in the series of dictionary registration modes ends. When the document correspondence dictionary is newly created, all the document feature information is newly registered in the dictionary feature database 10 as the dictionary feature information of the newly created document correspondence dictionary. When the document correspondence dictionary is not newly created, word information that does not overlap with both of the document feature information is added to the existing dictionary feature information of the document correspondence dictionary registered in the dictionary feature database 10. To be done. For example, when the document correspondence dictionary 17 is determined as the registration target, the dictionary feature information 302 illustrated in FIG. 3B includes “research” and “technology” in the document feature information 202 illustrated in FIG. , "And" information "are added.

By the operation in the dictionary registration mode as described above, the dictionary can be created and registered according to the characteristics of the document or additionally registered to the existing dictionary without the user being aware of the specialized field. The translated word for the same word can be arbitrarily registered for each such document corresponding dictionary. For example, as a translation of “structure”, “cons
It is possible to register "truction", "structure" in the document correspondence dictionary 17, and "organization" in the document correspondence dictionary 18.

FIG. 5 is a flow chart showing the operation of the translation mode in the machine translation device of the first embodiment.
Note that FIG. 6 shows an example of the document 601 to be translated, the characteristic information 602 of the document 601, and the translation result 603 obtained by translating the document 601.

When the process of the translation mode is started, the document input unit 7 takes in the document (601) to be translated input by the user (step 502). Then, the document feature extraction unit 9 extracts the document feature information (602) from the translation target document (step 503).

After that, the document correspondence dictionary determination unit 11 extracts the extracted document feature information (402) and the dictionary feature information (301, 302, 3 in the dictionary feature database 10).
03 ...) and the dictionary feature information having the maximum similarity is selected from the dictionary feature database 10 (step 504).

The dictionary name determined by the document correspondence dictionary determination unit 11 is passed to the translation interface unit 15 of the dictionary interface unit 4, and the translation interface unit 15 has the dictionary name in the dictionary unit 13 in the translation execution unit 3. The stored contents of the document corresponding dictionary are read (step 505). If there is no similar document corresponding dictionary, the stored contents of the general-purpose dictionary 19 having no special feature are read into the dictionary unit 13.

For example, in the range of information shown in FIGS. 3 and 6, the dictionary feature information 301 has a similarity of 1 to the document feature information 602 (“computers” are duplicated), and the dictionary feature information 302 is a document. The similarity to the feature information 602 is 4
(“Information”, “artificial intelligence”, “research”, and “processing” are duplicated), and the dictionary feature information 303 is the document feature information 602.
Is 0, and in this case, the document correspondence dictionary 17 is selected. That is, the stored contents of the document correspondence dictionary 17 are read into the dictionary unit 13.

If the document-corresponding dictionary does not store information that is not related to the document features such as particles and conjunctions, not only the stored contents of the determined document-corresponding dictionary,
The contents stored in the general-purpose dictionary 19 are transferred to the dictionary unit 13.

When the dictionary contents are transferred to the dictionary unit 13, the translation processing unit 12 uses this dictionary unit 13 to execute the translation process of the translation target document (601) (step 50).
6), the translation result output unit 8 outputs the obtained translation result (603).
Is output (step 507), and the processing in the series of translation modes ends. When the storage contents of the determined document-corresponding dictionary and the storage contents of the general-purpose dictionary 19 are transferred to the dictionary unit 13 and the source language word is redundantly described, the determined document-corresponding dictionary is stored. The stored contents have priority.

Here, each document corresponding dictionary 16, 17, 18
Is a translation of "structure",
It is assumed that “n”, “structure”, and “organization” are registered. When the translation target document 601 shown in FIG. 6A is input in such a situation, based on the document feature information 602, The document corresponding dictionary 17 having the most similar dictionary feature information is automatically selected, and the translation process is performed on the document-corresponding dictionary 17.
Is translated into

As described above, according to the first embodiment,
Instead of a fixed field unit according to the context vector,
It is possible to automatically create and automatically set the document correspondence dictionary in units according to the characteristics of the document.

Further, according to the first embodiment, the document feature is extracted from the document to be translated, and the optimum one is selected from a plurality of document correspondence dictionaries in units corresponding to the feature of the document. Even words can be translated according to the input document to be translated, and the translation quality can be further improved as compared with a conventional device that prepares a dictionary in fixed field units.

For example, in the dictionary selection method by field unit, both "structure: construction" and "structure: structure" are included in the same field of "information processing".
Although they are registered in the same dictionary and cannot be translated by selecting a dictionary, according to the first embodiment, since a document corresponding dictionary can be created for each feature unit of a document, each is registered in a different document corresponding dictionary. It is possible to perform translation by selecting a dictionary.

(B) Second Embodiment Next, a second embodiment in which the present invention is applied to a Japanese-English machine translation device will be described in detail with reference to the drawings. Here, FIG.
It is a functional block diagram which shows the machine translation apparatus of this 2nd Embodiment, Comprising: The same code | symbol is attached | subjected and shown to the same or corresponding part as FIG. 1 mentioned above.

In FIG. 7, the machine translation apparatus of the second embodiment is provided with a registration content extraction unit 20 in the dictionary determination unit 2. In the second embodiment, in the dictionary registration mode, a bilingual document of a source language Japanese document and its English document is input through the document input unit 7. The bilingual document input in this way is provided to the registered content extracting unit 20, and only the Japanese document of the entered bilingual documents is provided to the document feature extracting unit 9. .

The registered content extraction unit 20 functions only in the dictionary registration mode, and automatically extracts the content that can be registered in the dictionary from the bilingual documents composed of Japanese and English documents. As a method for automatically obtaining the contents that can be registered in the dictionary from the bilingual document, for example, the method described in Document 2 above can be applied.

The respective units other than the registered content extracting unit 20 function almost in the same manner as in the first embodiment. However, the dictionary creation unit 6
For example, among the registrable contents (word opposite information) extracted by the registration contents extracting unit 20, which are not stored in the document-corresponding dictionary to be registered, are presented to the user to determine whether or not registration is performed. Accepting content corrections, dictionary registration unit 14
In this way, the contents extracted by the registered contents extracting unit 20 instructed by the user via the dictionary creating unit 6 in this way are registered in a predetermined document corresponding dictionary.

Therefore, according to the second embodiment as well,
Instead of a fixed field unit, (1) it is possible to automatically create and automatically set the document correspondence dictionary in units according to the characteristics of the document, (2) extract the document characteristics from the translation target document, and Since the most suitable dictionary is selected from a plurality of document-corresponding dictionaries corresponding to the unit, it is possible to perform translation according to the input translation target document even for the same word,
It is possible to further improve the translation quality as compared with a conventional device that prepares a dictionary in fixed field units.

Further, according to the second embodiment, since the device side automatically obtains the contents that can be registered in the dictionary, it is possible to provide a useful device which can reduce the burden on the user.

(C) Other Embodiments In the description of each of the above embodiments, various modifications have been described, but other embodiments in which the above embodiment is modified as follows also constitute the present invention. To do.

(1) In each of the above-described embodiments, the document characteristic information and the dictionary characteristic information are mainly the group information of words that have appeared a predetermined number of times or more in the document, but they are other information. May be. For example, in order to reflect the length of the input document, the group information of words whose appearance rate obtained by dividing the number of appearances by the total number of words in the document is a predetermined appearance rate or more may be used as the document characteristic information and the dictionary characteristic information. good. Further, not only words but also idioms may be used as the constituent elements of the feature.

(2) Similarly, the similarity between the document feature information and the dictionary feature information is not limited to the number of words belonging to both. For example, the match / mismatch of the document creator may be converted into a value and included in the similarity value. For words (important words) having a large number of appearances or a high appearance rate, the value added to the degree of similarity may be increased.

(3) In each of the above embodiments, the present invention is applied to a Japanese-English machine translation device, but the present invention can be applied to a machine translation device whose source language or target language is other than this. Of course. Even in this case, it is necessary to obtain the document feature information and the dictionary feature information from the source language document.

(4) Furthermore, in each of the above embodiments,
Although the present invention is applied to a machine translation device having one translation direction, the present invention can be applied to a machine translation device having two or more translation directions. In this case, the dictionary feature database 10 needs to store the dictionary features in each language, and in the translation mode, it is necessary to obtain the document feature information from the original language document at that time.

(5) Furthermore, in each of the above embodiments, the dictionary feature information is stored in the dictionary feature database 10. However, the corresponding document corresponding dictionaries are provided with feature information storage areas. It may be stored.

(6) Further, in each of the above-described embodiments, one document-compatible dictionary provided for translation processing is shown.
Two or more document-corresponding dictionaries may be determined as those to be subjected to translation processing. In this case, the priority of use may be set according to the degree of similarity with the document feature of the translation target document.

[0071]

As described above, according to the dictionary registration apparatus of the first aspect of the present invention, the features of the input document are extracted to determine the necessity of creating the document correspondence dictionary, and if necessary, the document correspondence dictionary is created. Since it is created and registered, the document correspondence dictionary can be automatically created and automatically set in a unit according to the characteristics of the document, instead of a fixed field unit according to the context vector.

Further, according to the machine translation apparatus of the second aspect of the present invention, the document feature is extracted from the document to be translated, and the optimum one is selected from a plurality of document correspondence dictionaries in units corresponding to the feature of the document. Therefore, in addition to the effect of the dictionary registration device according to the first aspect of the present invention, even the same word can be translated according to the input translation target document, and the dictionary is prepared in fixed field units. It also has the effect of further improving the translation quality as compared with the device.

[Brief description of drawings]

FIG. 1 is a functional block diagram of a machine translation device according to a first embodiment.

FIG. 2 is an explanatory diagram of document characteristic information.

FIG. 3 is an explanatory diagram of dictionary feature information.

FIG. 4 is an operation flowchart in a dictionary registration mode according to the first embodiment.

FIG. 5 is an operation flowchart in a translation mode according to the first embodiment.

FIG. 6 is an explanatory diagram of a translation processing example according to the first embodiment.

FIG. 7 is a functional block diagram of a machine translation device according to a second embodiment.

[Explanation of symbols]

1 ... User interface part, 2 ... Dictionary determination part, 3 ...
Translation execution unit, 4 ... Dictionary interface unit, 5 ... Dictionary storage unit, 6 ... Dictionary creation unit, 9 ... Document feature extraction unit, 10 ... Dictionary feature database, 11 ... Document correspondence dictionary determination unit, 14
… Dictionary registration part, 15… Translation interface part, 16 ～
18 ... Document correspondence dictionary, 19 ... General-purpose dictionary.

Claims (3)

[Claims] 1. Creation of a dictionary used in a machine translation device,
In a dictionary registration device that performs registration, a document feature extraction unit that extracts document feature information including at least appearance frequency information of words and idioms in the input document from an input document described in natural language, and has already been created. For each of the one or more document-corresponding dictionaries, dictionary feature storage means for storing dictionary feature information having a format similar to the document feature information, document feature information extracted by the document feature extraction means, and the dictionary. The similarity with each dictionary feature information stored in the feature storage means is obtained, and it is necessary to create a new document correspondence dictionary based on each obtained similarity and the creation of a new document correspondence dictionary. When there is no need, the existing document correspondence dictionary to be registered is determined, and when a new document correspondence dictionary is created, the extracted document feature information is stored in the dictionary. Stored in the dictionary feature storage means as signature information, and when it is determined that the existing document correspondence dictionary is to be registered, it is stored in the dictionary feature storage means according to the extracted document feature information. Based on the similarity determining means for updating the dictionary feature information, and a new document correspondence dictionary is created if necessary based on the determination result of this similarity determining means, and then the dictionary is registered in the determined document correspondence dictionary. A dictionary registration device comprising: a dictionary registration means for performing. 2. A machine translation device having a dictionary registration structure, wherein first document characteristic information including at least appearance frequency information of words or idioms in the input document is extracted from the input document described in the source language. A document feature extraction unit, a dictionary feature storage unit that stores dictionary feature information having a format similar to the document feature information for each of the one or more document correspondence dictionaries already created, and the first document The similarity between the document feature information extracted by the feature extraction means and each dictionary feature information stored in the dictionary feature storage means is obtained, and a new document corresponding dictionary is created based on each obtained similarity. If there is no need to create a new document correspondence dictionary, the existing document correspondence dictionary to be registered is determined when there is no need to create a new document correspondence dictionary. The document feature information
The dictionary feature storage means stores the dictionary feature information, and when it is determined that the existing document corresponding dictionary is to be registered, the dictionary feature storage means stores the dictionary feature information according to the extracted document feature information. Based on the first similarity determination means for updating the existing dictionary feature information and the determination result of the first similarity determination means, if necessary, a new document correspondence dictionary is created, and then the determination is made. Dictionary registration means for registering a dictionary in the document-corresponding dictionary, and second document feature extraction means for extracting at least document feature information from the translation target input document, the appearance frequency information of words and idioms in the translation target input document. And a similarity between the document feature information extracted by the second document feature extraction means and each dictionary feature information stored in the dictionary feature storage means, and based on each obtained similarity. , Second similarity determination means for determining one or more document correspondence dictionaries to be used for translation processing, and the contents of the document correspondence dictionary determined to be used for translation processing for the translation target input document. A machine translation device, comprising: 3. The same document feature extracting means is selectively used as the first and second document feature extracting means, and the same similarity determining means is selectively used as the first and second similarity determining means. The machine translation device according to claim 2.