JP3419748B2 - Dictionary creation device and method, and recording medium recording dictionary creation program - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dictionary creating apparatus that is used in, for example, a natural language processing system, an information search system, a machine translation system, etc., to create a bilingual dictionary for different natural language pairs.

[0002]

2. Description of the Related Art In recent years, with the development of the Internet and the like, opportunities to come into contact with documents written in various natural languages have increased, and users have also increased opportunities to come into contact with natural languages that they are not necessarily fluent in. Along with this, because users use documents and information written in natural language that they are not always fluent in, multilingual search and multilingual information processing, especially for electronic data,
The demand for multilingual translation and the like has increased, and a dictionary of various language pairs for use in them has been required.
Therefore, it is required to create high-quality bilingual dictionaries for various language pairs.

It is common to create a high-quality dictionary manually (Shirai et al., "Number of combined pattern pairs necessary for Japanese-English machine translation and its collection method", IPSJ research report, Nature Language Processing Research Group, 95-NL-110-7, pp.43-50), because the amount of work is enormous, it takes a lot of time, labor and time to create a dictionary manually. In particular, in recent years, extension to various language pairs has been required, and therefore it is desired to automatically or semi-automatically create a high-quality dictionary without human intervention.

As a method of automatically creating a dictionary of natural language pairs, there is a method of utilizing a bilingual corpus of natural language pairs to be created. A bilingual corpus is an electronic version of the same document written in different languages. The method of using the bilingual corpus is to morphologically analyze the bilingual corpus of natural language pairs with sentence correspondence, and find the appearance frequency in all bilingual corpus for all word pairs of language pairs appearing in the bilingual sentence. , Word pairs with high frequency of occurrence are calculated to have a high degree of similarity, the word similarity is calculated, and the bilingual corpus is syntactically analyzed to compare the syntactic structure, and the translation pattern is extracted from the comparison between the word similarity and the syntactic structure. (Kitamura et al., "Automatic acquisition of translation knowledge from bilingual bilingual corpus", Proceedings of the 8th National Congress of the Japanese Society for Artificial Intelligence, pp.645-648, 1994).

However, although a large number of corpora have become available on computers in recent years, there are few types and quantities of bilingual corpora, and a large-scale dictionary that can withstand practical use of bilingual corpora of various language pairs. Hard to get enough to make (KW Church and R.L. Mercer, "Introduction
to the special issue on computational linguistics
using large corpora ", Computational Linguistics Vo
l.19, PP1-24, 1993). In particular, there are fewer bilingual corpora with sentence correspondence, and bilingual corpuses without sentence correspondence do not always have one-to-one correspondence at the sentence level, or they do not appear as accurate translations, such as free translation. .

[0006]

Among the above-mentioned conventional techniques, the manual method requires too much cost, manpower and time as compared with the diversification of the required dictionary language pairs and the increase in needs. There is a problem.

Further, in the method using the bilingual corpus,
The bilingual corpus is small in type and amount, and there is a problem that it is difficult to obtain a bilingual corpus of a sufficient amount to create a large-scale dictionary that can be practically used for various language pairs.

The present invention has been made in view of the above,
It is an object of the present invention to provide a dictionary creating device and method capable of accurately creating a dictionary between high-quality natural languages without a sufficient amount of bilingual corpus, and a recording medium recording the dictionary creating program. It is in.

[0009]

In order to achieve the above object, the present invention according to claim 1 is a dictionary creating apparatus for creating a bilingual dictionary for natural language pairs, wherein a headline of the first natural language and its part of speech are provided. And a first natural language to second natural language dictionary storing a first natural language to second natural language dictionary having a second natural language bilingual expression which is a parallel translation of the first natural language entry and its part of speech. A second natural language vs. third natural language dictionary having a part, a headline of the second natural language and its part of speech, a parallel translation of the third natural language which is a parallel translation of the second natural language entry, and its part of speech is stored. A second natural language vs. third natural language dictionary storage, and the first natural language vs. second
A dictionary entry reading unit that extracts a first natural language-to-second natural language dictionary entry from the natural language dictionary storage unit and a third natural language entry corresponding to the second natural language entry of the extracted first natural language-to-second natural language dictionary entry. Creating a dictionary in which a natural language entry is obtained from the second natural language to third natural language dictionary storage unit and the second natural language entry of the first natural language to second natural language dictionary entry is replaced with the third natural language entry And a first natural language vs. third natural language dictionary entry generated by replacing the second natural language entry of the first natural language entry vs. second natural language dictionary entry with the third natural language entry. 1
Creating and storing a first natural language-to-third natural language bilingual dictionary having a natural language heading and its part of speech, and a third natural language bilingual translation and its part of speech, which are parallel translations of the first natural language entry. It has a first natural language-to-third natural language dictionary storage unit and a first natural language corpus information storage unit that stores statistical information of the first natural language corpus, and the dictionary creation management unit is the dictionary entry reading unit. The first natural language to second natural language dictionary entry supplied from the first natural language entry is used as a key to refer to the first natural language corpus information storage unit, and unnecessary first natural language to second natural language The gist is to have an entry selection information addition unit for removing entries.

According to the first aspect of the present invention, the first natural language pair second natural language dictionary entry is retrieved from the first natural language pair second natural language dictionary storage unit, and the retrieved first natural language pair is retrieved. A third natural language entry corresponding to the second natural language entry of the second natural language dictionary entry is obtained from the second natural language-to-third natural language dictionary storage unit, and the third natural language entry is used as the first natural language-to-first natural language entry. A first natural having a headline and a part of speech of the first natural language, replacing the second natural language entry of the second natural language dictionary entry, and a parallel translation of a third natural language which is a parallel translation of the first natural language entry and its part of speech. First to create a bilingual dictionary for language vs. third natural language
The first natural language to the third natural language dictionary is stored by using the first natural language to the second natural language dictionary and the second natural language to the third natural language dictionary for storing in the natural language to the third natural language dictionary storage unit. Even if there is no dictionary or bilingual corpus of combinations of
A natural language to a third natural language dictionary can be created.
Also, the first natural language corpus information storage unit is referred to by using the first natural language entry of the first natural language pair second natural language dictionary entry supplied from the dictionary entry reading unit as a key, and unnecessary first natural language pair Since the second natural language entry is removed, unnecessary dictionary entries are not created when, for example, creating a domain-specific dictionary.
As a result, speeding up and efficiency can be achieved in creating and using the first natural language-to-third natural language dictionary. Further, when a dictionary limited to a field is created, if the first natural language vs. second natural language dictionary and the second natural language vs. third natural language dictionary to be used can also use the dictionary of the field, it is more reliable. It is possible to create a first natural language-to-third natural language dictionary that is field limited.

Further, the present invention according to claim 2 provides the invention according to claim 1.
In the invention described above, the method further comprises a second natural language corpus information storage unit that stores statistical information of a second natural language corpus, and the entry selection information adding unit is the first natural language corpus information supplied from the dictionary entry reading unit. Translated word selection information / weighted information addition, which refers to the second natural language corpus information storage unit using the second natural language entry of the natural language vs. second natural language dictionary entry as a key, and adds translated word selection information that is weighting information The gist is to have means.

According to the second aspect of the present invention, the second natural language corpus information is stored by using the second natural language entry of the first natural language to second natural language dictionary entry supplied from the dictionary entry reading section as a key. Even if there are a plurality of translation word candidates from the first natural language to the second natural language, for example, as a technical term or a field-specific word, which translation word Is more appropriate, and among multiple candidate words,
It is possible to prioritize which translation word candidate appears more often. Furthermore, even if there is only one translation word candidate, the likelihood of the translation word candidate can be determined, and as a result,
A more reliable first natural language to third natural language dictionary can be created. Further, since the information for selecting the translated word is added at the stage of the second natural language, the same information can be used when changing only the third natural language without changing the first natural language and the second natural language. As a result, when the third natural language is expanded into various languages, the first natural language-to-third natural language dictionary can be efficiently created.

According to a third aspect of the present invention, in the first or second aspect of the invention, the dictionary creation management unit sets the second natural language entry of the first natural language to second natural language dictionary entry to the third natural language entry. If it cannot be replaced with the language entry, a second natural language analysis unit that analyzes the second natural language entry and a third natural language entry is obtained based on the analysis result of the second natural language analysis unit. 3rd won
The gist is to have a replacement execution unit that replaces the second natural language entry with a natural language entry.

According to the third aspect of the present invention, when the second natural language entry of the first natural language to second natural language dictionary entry cannot be replaced with the third natural language entry, the second natural language entry is replaced with the second natural language entry. The natural language entry is analyzed, a third natural language entry is acquired based on the analysis result, and the second natural language entry is replaced with the acquired third natural language entry. For example, the second natural language entry is a compound word, Even if it is not registered as it is in the second natural language vs. third natural language dictionary, the second natural language entry is analyzed by dividing it into sub-expression units constituting a compound word, and each sub-expression is divided. Is registered in the second natural language to third natural language dictionary, the second natural language entry of the first natural language to second natural language dictionary entry is replaced with the third natural language entry. Rukoto can, can be created as a result, as it is the third natural language entry also won the first natural language versus the third natural language dictionary for such difficult compound word of the replacement notation.

The present invention according to claim 4 is the same as claim 3
In the invention described above, the method further includes a replacement rule storage unit that stores a replacement rule for replacing the second natural language with the third natural language, and the dictionary creation management unit stores the analysis result of the second natural language analysis unit. And a replacement rule retrieval unit that retrieves a replacement rule based on the replacement rule storage unit based on the replacement rule retrieval unit. Has a means for acquiring a third natural language entry from the second natural language vs. third natural language dictionary storage unit by using the plurality of replacement rules in combination.

According to the present invention of claim 4, the replacement rule storage section is searched based on the analysis result of the second natural language analysis section, the replacement rule is retrieved, and the retrieved replacement rule is used. When there are a plurality of replacement rules, the plurality of replacement rules are used in combination, and the second natural language vs. the third
In order to obtain the third natural language entry from the natural language dictionary storage unit, for example, the second natural language entry is a compound word, an inflected form, or a combination thereof.
Even if it is not registered in the natural language-to-third natural language dictionary, the analysis result of the second natural language entry, the substitution rule suitable for the analysis result, and the second natural language-to-third natural language dictionary are used to make it more complicated. The second natural language dictionary entry of the first natural language to the second natural language dictionary entry more reliably
The natural language entry can be replaced with a third natural language entry. In addition, even if the replacement rule that can replace the analysis result of the second natural language entry in a lump by using a plurality of replacement rules in combination is not stored in the replacement rule storage unit, the first natural language vs. second natural language The second natural language entry of the natural language dictionary entry is the third
A third natural language entry for a more complicated second natural language entry, such as a compound word that can be replaced by a natural language entry, and as a result is difficult to replace in its original form, is accurately acquired, and the first natural language vs. 3 Can create a natural language dictionary. Furthermore, since the replacement rules can be used in combination, it is not necessary to store all possible replacement rules in the replacement rule storage unit, and the first natural language-to-third natural language dictionary can be efficiently created. it can.

Furthermore, the present invention according to claim 5 provides the invention according to claim 4.
In the invention described above, when the dictionary creation management unit uses a plurality of replacement rules retrieved by the replacement rule search unit in the replacement execution unit in combination, the number of times of use of counting the number of times of use of the combined replacement rules is counted. The gist of the present invention is to have means and a replacement rule creating unit for storing the combined replacement rule in the replacement rule storage unit when the counted number of times of use of the replacement rule exceeds a predetermined number.

In a fifth aspect of the present invention, when a plurality of replacement rules are used in combination, the number of times of use of the combined replacement rules is counted, and when the number of times of use exceeds a predetermined number of times, Since this combined replacement rule is stored in the replacement rule storage unit, a combination replacement rule that is frequently used is acquired and stored in the replacement rule storage unit, so that the first natural language versus the second natural language dictionary entry is stored. It is possible to speed up when replacing the second natural language entry with the third natural language entry, and it is possible to efficiently create the first natural language-to-third natural language dictionary.

The present invention according to claim 6 is based on claims 3 to 5.
In any one of the inventions, the second natural language dictionary storage unit for storing the second natural language dictionary storing the original form, the part of speech, and the utilization information of the second natural language is further provided. The gist is that the analysis unit has a unit that searches the second natural language dictionary storage unit and analyzes the second natural language entry.

According to a sixth aspect of the present invention, the second natural language stores the original form, part of speech, and utilization information.
It has a natural language dictionary, and the second natural language analysis unit searches the second natural language dictionary storage unit to analyze the second natural language entry. Even if it is not registered in the second natural language-to-third natural language dictionary, the original form of the variation included in the second natural language entry can be obtained from the second natural language dictionary. When registered in the language-to-third natural language dictionary, the second natural language entry of the first natural language-to-second natural language dictionary entry can be replaced with the third natural language entry, and as a result, the notation is replaced. It is also possible to obtain a third natural language entry for the second natural language entry such as a variation that is difficult to perform, and it is possible to accurately create the first natural language-to-third natural language dictionary.

The present invention according to claim 7 provides the invention according to claim 3.
The invention according to any one of claims 1 to 6, further comprising a third natural language dictionary storage unit that stores a third natural language dictionary that stores an original form, a part of speech, and utilization information of the third natural language,
The replacement execution unit searches the third natural language dictionary storage unit based on the analysis result of the second natural language analysis unit or the replacement rule retrieved by the replacement rule search unit to determine whether the first natural language versus the second natural language. The gist is to have a means for replacing the second natural language entry of the language dictionary entry with the third natural language entry.

According to a seventh aspect of the present invention, a third natural language, which stores an original form, a part of speech, and utilization information, is stored.
A natural language dictionary is provided, and the third natural language dictionary storage unit is searched based on the analysis result of the second natural language analysis unit or the replacement rule extracted by the replacement rule search unit, and the first natural language versus the second natural language. Since the second natural language entry of the dictionary entry is replaced with the third natural language entry, for example, the second natural language entry contains a variation and is not registered in the second natural language vs. third natural language dictionary as it is. In this case, when the variation included in the second natural language entry is corrected to the original form and the corresponding third natural language entry is obtained in the original form from the second natural language-to-third natural language dictionary, the third natural language dictionary To convert the original form of the third natural language entry into an appropriate variant notation, and as a result, it corresponds to the second natural language entry such as a variant that is difficult to replace with the original notation. 3 is converted into a natural language entry is also applicable exact representation can be created accurately first natural-language third natural language dictionary.

Furthermore, the present invention according to claim 8 is directed to claim 1.
The invention according to any one of claims 1 to 7, wherein the dictionary creation management unit obtains a plurality of second natural language entries of the first natural language vs. second natural language dictionary entry, and sets the third natural language entry to the third natural language entry. In the case where the natural language entries are acquired redundantly, it is a gist to have a translation word selection information addition unit that determines that the third natural language entry acquired redundantly is more appropriate.

According to the present invention of claim 8, a plurality of second natural language entries of the first natural language vs. second natural language dictionary entry are obtained, and the third natural language entry for the second natural language entry is obtained. Is duplicated,
In order to determine that the third natural language entry obtained by duplication is more appropriate, if there are a plurality of translation candidates at the stage of the second natural language entry, it is obtained from a plurality of second natural language entries. If the duplicated third natural language entries are duplicated, it is possible to add translation word selection information that the duplicated third natural language entry is more appropriate than the third natural language entry that has appeared independently, and as a result, it is possible to improve the reliability. A high first natural language to third natural language dictionary can be created.

The present invention according to claim 9 relates to claims 1 to 8.
In any one of the inventions described above, the method further comprises a third natural language corpus information storage unit that stores statistical information of a third natural language corpus, and the dictionary creation management unit stores the third natural language corpus information storage unit. It is a gist to have means for adding translation word selection information indicating whether or not the third natural language entry acquired by the dictionary creation management unit is appropriate using the information of the copy.

According to the present invention of claim 9, there is provided a third natural language corpus information storage section for storing statistical information of the third natural language corpus, and the third natural language corpus information storage section stores the third natural language corpus information storage section. Using the information, in order to add translated word selection information indicating whether or not the third natural language entry acquired by the dictionary creation management unit is appropriate, the first natural language entry through the second natural language entry to the third natural language entry Even if the number of translation candidates increases while following the entry, it is possible to know which translation word is more appropriate, for example, a technical term or a field-specific word. Also,
Among the plurality of translation word candidates, the more frequently used word can be given priority, and the likelihood of the translation word candidate can be determined even when there is only one translation word candidate in the third natural language.
Furthermore, when the second natural language entry is a compound word and the third natural language entry cannot be acquired by the notation as it is, and the analysis is performed to acquire the third natural language entry, word candidates in the third natural language entry Even if there is more than one, the third won
A more appropriate combination of words can be selected from co-occurrence information between words included in the natural language entry, and as a result, a highly reliable first natural language versus third natural language dictionary can be created. .

The present invention according to claim 10 is a dictionary creating method for creating a bilingual dictionary for a natural language pair, which is a bilingual translation of a headline of the first natural language and its part of speech, and the first natural language entry. First natural language to second natural language dictionary storage for storing a first natural language to second natural language dictionary having two natural language bilingual expressions and their parts of speech in a first natural language to second natural language dictionary storage unit A second natural language-to-third natural language dictionary having a part storing step, a headline of the second natural language and its part of speech, a parallel translation of a third natural language which is a parallel translation of the second natural language entry, and its part of speech. A step of storing a second natural language pair to a third natural language dictionary storage section for storing the second natural language pair to the third natural language dictionary storage section; and a first natural language pair from the first natural language to second natural language dictionary storage section. Second
A dictionary entry reading step for fetching a natural language dictionary entry, and a third natural language entry corresponding to the second natural language entry of the fetched first natural language vs. second natural language dictionary entry.
Creating a dictionary in which a natural language entry is obtained from the second natural language to third natural language dictionary storage unit and the second natural language entry of the first natural language to second natural language dictionary entry is replaced with the third natural language entry A first step based on the managing step and the first natural language to third natural language dictionary entry generated by replacing the second natural language entry of the first natural language to second natural language dictionary entry with the third natural language entry; Creating and storing a first natural language-to-third natural language bilingual dictionary having a natural language heading and its part of speech, and a third natural language bilingual translation and its part of speech, which are parallel translations of the first natural language entry. 1 natural language vs. 3rd natural language dictionary storing step, and storing the statistical information of the 1st natural language corpus in the 1st natural language corpus information storing section, The dictionary creation management step refers to the first natural language corpus information storage unit using the first natural language entry of the first natural language to second natural language dictionary entry supplied from the dictionary entry reading step as a key, and is unnecessary. The gist of the present invention is to have an entry selection information providing step for removing the first natural language vs. second natural language entry.

According to the present invention of claim 10,
A first natural language-to-second natural language dictionary entry is retrieved from the natural language-to-second natural language dictionary storage unit, and a third natural language entry corresponding to the retrieved second natural language entry of the first natural language-to-second natural language dictionary entry is retrieved. A natural language entry is obtained from a second natural language to third natural language dictionary storage, and the third natural language entry replaces the second natural language entry of the first natural language to second natural language dictionary entry to create a first natural language entry. A first natural language bilingual dictionary is created by creating a bilingual dictionary of a first natural language to a third natural language, which has a headline of the language and its part of speech, and a bilingual translation of the third natural language which is a bilingual translation of the first natural language entry and its part of speech Since it is stored in the storage unit for the third natural language dictionary,
Using the first natural language-to-second natural language dictionary and the second natural language-to-third natural language dictionary, it is possible to use the first natural language without the dictionary or bilingual corpus of the combination of the first natural language and the third natural language. A language-to-third natural language dictionary can be created. Further, the first natural language corpus information storage unit is referred to by using the first natural language entry of the first natural language to second natural language dictionary entry supplied from the dictionary entry reading step as a key, and unnecessary first natural language In order to remove the second-to-second natural language entry, unnecessary dictionary entries will not be created, for example, when creating a domain-specific dictionary, resulting in the creation of the first-to-third natural language-to-third natural language dictionary. It is possible to increase the speed and efficiency of use. Further, when a dictionary limited to a field is created, if the first natural language vs. second natural language dictionary and the second natural language vs. third natural language dictionary to be used can also use the dictionary of the field, it is more reliable. Field-limited first natural language vs. third
A natural language dictionary can be created.

Further, the invention according to claim 11 is the invention according to claim 10, further comprising a step of storing statistical information of the second natural language corpus in a second natural language corpus information storage section, and the entry. The selection information giving step refers to the second natural language corpus information storage unit with the second natural language entry of the first natural language vs. second natural language dictionary entry supplied from the dictionary entry reading step as a key, and weights it. The gist is to have a translation word selection information / weighting information addition step of adding translation word selection information that is information.

According to the present invention of claim 11, the second natural language corpus information is stored by using the second natural language entry of the first natural language vs. second natural language dictionary entry supplied from the dictionary entry reading step as a key. See the section
In order to add translation word selection information which is weighting information, the first
Even when there are multiple candidate translations from a natural language to a second natural language, it is possible to know which translation is more appropriate, for example, as a technical term or field-specific word, and it appears more often among multiple candidate translations. It is possible to prioritize which of the translated word candidates to be processed is. Further, even when there is only one translation word candidate, the likelihood of the translation word candidate can be determined, and as a result, a more reliable first natural language versus third natural language dictionary can be created. Further, since the information for selecting the translated word is added at the stage of the second natural language, the same information can be used when changing only the third natural language without changing the first natural language and the second natural language. As a result, when the third natural language is expanded into various languages, the first natural language-to-third natural language dictionary can be efficiently created.

According to a twelfth aspect of the present invention, in the invention according to the tenth or eleventh aspect, the dictionary creation management step is the second natural language dictionary entry of the second natural language dictionary entry.
If the natural language entry cannot be replaced with the third natural language entry, the second natural language analysis step of analyzing the second natural language entry and the third natural language based on the analysis result of the second natural language analysis step. The gist is to have a replacement execution step of acquiring a language entry and replacing the second natural language entry with the acquired third natural language entry.

According to the twelfth aspect of the present invention, there is provided the first aspect.
If the second natural language entry of the natural language vs. second natural language dictionary entry cannot be replaced with the third natural language entry, the second natural language entry is analyzed and the third natural language entry is analyzed based on this analysis result. In order to replace the second natural language entry with the acquired third natural language entry, for example, the second natural language entry is a compound word and is registered in the second natural language vs. third natural language dictionary as it is. Even if not, if the second natural language entry is analyzed by dividing it into sub-expression units that form a compound word, and if each sub-expression is registered in the second natural language to third natural language dictionary, , The second natural language entry of the first natural language vs. second natural language dictionary entry can be replaced with the third natural language entry, resulting in the replacement of the natural notation. The third natural language entry for such difficult compound words can create the first natural language versus the third natural language dictionary won.

According to a thirteenth aspect of the present invention, there is provided a recording medium in which a dictionary creating program for creating a bilingual dictionary for a natural language pair is recorded, the headline of the first natural language and its part of speech, and the first natural language entry. A first natural language to second natural language dictionary having a second natural language bilingual expression and its part-of-speech that are parallel translations of the first natural language to second natural language dictionary storing section; A second natural language dictionary storing step, a second natural language headline and its part of speech, a third natural language bilingual expression that is a parallel translation of the second natural language entry, and its part of speech; From the second natural language-to-third natural language dictionary storage section storing step of storing the third natural language dictionary in the second natural language-to-third natural language dictionary storage section, and from the first natural language-to-second natural language dictionary storage section. First natural language A dictionary entry reading step of fetching a pair of second natural language dictionary entries, and a third natural language entry corresponding to the second natural language entry of the fetched first natural language pair second natural language dictionary entry is set as the second natural language pair. A dictionary creation management step of replacing the second natural language entry of the first natural language pair second natural language dictionary entry with the third natural language entry acquired from the third natural language dictionary storage, and the first natural language pair A headline of the first natural language and its part of speech based on the first natural language vs. the third natural language dictionary entry generated by replacing the second natural language entry of the second natural language dictionary entry with the third natural language entry; A first natural language to a third natural language bilingual dictionary having a third natural language bilingual translation that is a bilingual translation of the first natural language entry and its part of speech A first natural-language third natural language dictionary storage step of storing create and to,
Storing statistical information of the first natural language corpus in a first natural language corpus information storage unit, wherein the dictionary creation management step includes the first natural language vs. second natural language supplied from the dictionary entry reading step. An entry selection information adding step for removing unnecessary first natural language vs. second natural language entries by referring to the first natural language corpus information storage unit using the first natural language entry of the language dictionary entry as a key The main point is to record the dictionary creation program.

According to the invention of claim 15, the first aspect is
A first natural language-to-second natural language dictionary entry is retrieved from the natural language-to-second natural language dictionary storage unit, and a third natural language entry corresponding to the retrieved second natural language entry of the first natural language-to-second natural language dictionary entry is retrieved. A natural language entry is obtained from a second natural language to third natural language dictionary storage, and the third natural language entry replaces the second natural language entry of the first natural language to second natural language dictionary entry to create a first natural language entry. A first natural language bilingual dictionary is created by creating a bilingual dictionary of a first natural language to a third natural language, which has a headline of the language and its part of speech, and a bilingual translation of the third natural language which is a bilingual translation of the first natural language entry and its part of speech. The first natural language corpus information storage unit is referred to with the first natural language entry of the supplied first natural language to second natural language dictionary entry stored as a key in the third natural language dictionary storage unit. , Because it records a dictionary creation program to remove the first natural-language second natural language entry unnecessary in the recording medium, by using the recording medium, it is possible to enhance the flow properties.

[0035]

[0036]

[0037]

[0038]

[0039]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of a dictionary creating apparatus according to an embodiment of the present invention. The dictionary creating apparatus shown in the figure includes a first natural language vs. second natural language dictionary storage unit 10, a dictionary entry reading unit 11, a dictionary creation managing unit 12, a first natural language vs. third natural language dictionary storage unit 13,
A first natural language corpus information storage unit 14, a second natural language corpus information storage unit 15, a replacement rule storage unit 16, a second natural language dictionary storage unit 17, a third natural language dictionary storage unit 18,
It is composed of a second natural language-to-third natural language dictionary storage unit 19 and a third natural language corpus information storage unit 20.

First natural language vs. second natural language dictionary storage unit 1
0 stores one or more first natural language-to-second natural language dictionaries, and at least one heading of the first natural language and its part of speech and the first natural language are stored in the stored dictionary records. And the part of speech of the second natural language corresponding to the first natural language entry which is a pair of the heading of and the part of speech.

The dictionary entry reading unit 11 includes a second natural language entry from the dictionary stored in the first natural language vs. second natural language dictionary storage unit 10 and a second natural language entry including the bilingual expression.
The natural language entry is extracted and supplied to the dictionary creation management unit 12.

The dictionary creation management unit 12 includes a first natural language corpus information storage unit 14, a second natural language corpus information storage unit 15, a replacement rule storage unit 16, a second natural language dictionary storage unit 17, and a third natural language dictionary. The first natural language vs. second natural language dictionary supplied from the dictionary entry reading unit 11 with reference to the storage unit 18, the second natural language vs. third natural language dictionary storage unit 19, and the third natural language corpus information storage unit 20. The second natural language entry of the entry is replaced with the third natural language entry and stored in the first natural language vs. third natural language dictionary storage unit 13.

The dictionary creation management unit 12 comprises an entry selection information addition unit 121, a second natural language analysis unit 122, a substitution management unit 123, and a translation word selection information addition unit 124. In addition, the replacement management unit 123 includes a replacement rule search unit 1231,
The replacement rule creation unit 1232 and the replacement execution unit 1233 are included.

The entry selection information addition unit 121 of the dictionary creation management unit 12 uses the first natural language entry of the first natural language vs. second natural language dictionary entry supplied from the dictionary entry reading unit 11 as a key, and sets the first natural language entry as a key. By referring to the language corpus information storage unit 14, it is determined whether the first natural language to second natural language dictionary entry is necessary, and if it is unnecessary, the subsequent dictionary entry reading unit 1 is performed without performing the subsequent operation.
The operation is moved to the first natural language to second natural language dictionary entry supplied from 1.

Further, the entry selection information adding section 121
The second natural language corpus information storage unit 15 is referred to by using the second natural language entry of the first natural language to second natural language dictionary entry supplied from the dictionary entry reading unit 11 as a key, and related to the second natural language entry. The statistical information in the corpus is taken out and added to the second natural language entry, and the replacement executing unit 1233 in the replacement managing unit 123 of the dictionary creation managing unit 12 is supplied with the second natural language entry to which the statistical information is added.

The replacement executing unit 1233 is the second natural language vs. the third natural language.
By referring to the natural language dictionary storage unit 19, the second natural language entry of the first natural language versus second natural language dictionary entry supplied from the dictionary entry reading unit 11 is replaced with the third natural language entry, and the first natural language entry And a set of third natural language entries are supplied to the translated word selection information addition unit 124.

However, when the replacement execution unit 1233 cannot acquire the third natural language entry for the second natural language entry of the first natural language to second natural language dictionary entry, the second natural language entry is managed to create a dictionary. It is supplied to the second natural language analysis unit 122 of the unit 12.

The second natural language analysis section 122 of the dictionary creation management section 12 has a first analysis means for the second natural language entry.
One or more, and the substitution execution unit 1233 of the substitution management unit 123 sets the second natural language to second natural language dictionary entry
When the natural language entry cannot be replaced with the third natural language entry, the second natural language dictionary storage unit 17 is referenced as needed to analyze the second natural language entry, and the analysis result is replaced and managed. It is supplied to the replacement rule search unit 1231 of the unit 123.

The substitution management unit 123 of the dictionary creation management unit 12
Obtains a third natural language entry for the second natural language dictionary entry of the first natural language versus second natural language dictionary entry.

Replacement rule search unit 12 of replacement management unit 123
31 is the first supplied from the second natural language analysis unit 122.
The replacement rule of the replacement rule storage unit 16 is searched using the analysis result of the second natural language entry of the natural language to second natural language dictionary entry as a key, and the second natural language entry of the first natural language to second natural language dictionary entry is searched. One or more replacement rules for replacing the second natural language entry with the third natural language entry are extracted by using the analysis result of 1.

The replacement rule creation unit 12 of the replacement management unit 123
32 is a combination of a plurality of existing replacement rules stored in the replacement rule storage unit 16 in the replacement rule search unit 1231, and a second natural language dictionary second second natural language dictionary entry.
When the analysis result for the natural language entry is replaced and the number of appearances of the combination of the replacement rules (combination replacement rule) exceeds a certain number, the combination replacement rule is stored in the replacement rule storage unit 16.

Replacement execution unit 1233 of replacement management unit 123
Searches the second natural language-to-third natural language dictionary storage unit 19 using the second natural language entry of the first natural language-to-second natural language dictionary entry as a key, and searches the first natural language-to-second natural language dictionary. Replace the second natural language entry of the entry with the third natural language entry.

The substitution executing unit 1233 also analyzes the result of the second natural language analysis unit 122 for the second natural language entry of the first natural language vs. second natural language dictionary entry, or analyzes the second natural language entry. According to the result and the one or more replacement rules retrieved by the replacement rule search unit 1231, the second natural language vs. third natural language dictionary storage unit 19
Then, the third natural language dictionary storage unit 18 is referred to when necessary, and the second natural language entry of the first natural language vs. second natural language dictionary entry is replaced with the third natural language entry.

Further, when the substitution executing unit 1233 cannot acquire the third natural language entry corresponding to the second natural language entry of the first natural language vs. second natural language dictionary entry, the second natural language entry is obtained. The second natural language analysis unit 1
Returning to step 22, the process of replacing the second natural language entry of the first natural language vs. second natural language dictionary entry with the third natural language entry is repeated in the same manner as above.

However, the second natural language analysis unit 122 returns the second natural language entry of the first natural language to second natural language dictionary entry to the second natural language analysis unit 122 even once. Only when the second natural language analysis unit 122 stores a plurality of analysis methods and there is an analysis method that has not been performed for the second natural language entry. If not, a flag indicating that the replacement could not be performed is added, and the second natural language entry of the first natural language vs. second natural language dictionary entry is not replaced with the third natural language entry, or the second natural language entry is not replaced. The result of the replacement of the natural language entry with the third natural language entry is stored in the first natural language-to-third natural language dictionary storage unit 13.

The translation word selection information adding unit 124 refers to the third natural language corpus information storage unit 20 with the third natural language entry supplied from the substitution managing unit 123 as a key, and refers to each of the third natural language entries. The translation word selection information such as which translation word to prioritize or select is added.

First natural language to third natural language dictionary storage unit 1
Reference numeral 3 denotes a first natural language heading and its part of speech supplied from the dictionary creation management unit 12 and a first natural language entry corresponding to the first natural language heading and its part of speech pair. It is also possible to store the bilingual expression of the three natural languages, its part of speech, and, if necessary, the translation selection information of the third natural language.

The first natural language corpus information storage unit 14 is
The statistical information extracted using the first natural language corpus, which is an electronic document, is stored.

The second natural language corpus information storage unit 15 stores
The statistical information extracted using the second natural language corpus, which is an electronic document, is stored.

The replacement rule storage unit 16 includes a replacement management unit 12
3, the replacement rules used when replacing the second natural language entry of the first natural language vs. second natural language dictionary entry with the third natural language entry are stored.

Further, the replacement rule storage unit 16 also stores the combination replacement rules that have been used a certain number of times among the combination replacement rules used by combining the existing replacement rules in the replacement rule search unit 1231.

The second natural language dictionary storage unit 17 stores a monolingual dictionary of the second natural language, and stores at least the original form, part of speech, and utilization information as dictionary entries.

The third natural language dictionary storage section 18 stores a monolingual dictionary of the third natural language, and stores at least the original form, part of speech, and utilization information as dictionary entries.

Second natural language vs. third natural language dictionary storage unit 1
9 stores one or more second natural language-to-third natural language dictionaries, and the stored second natural language-to-third natural language dictionary entry stores at least the headline of the second natural language and its part of speech. And a third corresponding to the second natural language entry
It contains natural language bilingual expressions and their parts of speech.

The third natural language corpus information storage unit 20 is
The statistical information extracted using the third natural language corpus, which is an electronic document, is stored.

Next, a specific example and operation of the dictionary creating apparatus configured as described above will be described with reference to the flow chart shown in FIG. In this description, the first natural language is Japanese, the second natural language is English, and the third natural language is French. However, the combination of the first natural language, the second natural language, and the third natural language is Any language such as Thai, Chinese, Spanish, German, French, Russian, etc. may be used as long as the first, second, and third natural languages are different.

FIG. 2 shows the first natural language vs. the second natural language shown in FIG.
5 shows an example of first natural language versus second natural language dictionary entries stored in the natural language dictionary storage unit 10. The first natural language-to-second natural language dictionary has the first natural language as Japanese and the second natural language.
If the natural language is English, for example, ALT-J / E's Japanese-English contrast dictionary (NTT, "Japanese-English machine translation technology", NTTR &
Dvol.46, pp107-141,1997) and EDR Electronic Dictionary (Nihondenshikajishokenkyusho, http: //www.iijnet.or.jp/edr/J _- in
dex.html) exists.

As shown in FIG. 2, as the first natural language versus second natural language dictionary entry, at least the Japanese headline and its part of speech, and the English bilingual expression for the Japanese entry and its part of speech are registered. For example, the first natural language vs. second natural language dictionary entry whose Japanese heading is "bank" has its part of speech "noun" and its bilingual expression "bank" and its part of speech "noun" registered. The first natural language vs. second natural language dictionary entry with the word heading "bank clerk" has its part of speech "noun" and bilingual expression "bank cle
rk ”and its part-of-speech“ noun ”are registered, and the first natural language vs. second natural language dictionary entry whose Japanese heading is“ Yawake ”is the part-of-speech“ noun ”and bilingual expression”
burn ”and its part-of-speech“ noun ”, and yet another parallel translation“ scald ”and its part-of-speech“ noun ”are registered, and the Japanese heading is“ Four Seasons ”. In the natural language dictionary entry, the part-of-speech "noun", the bilingual expression "four seasons", and the part-of-speech "noun" are registered. Furthermore, the Japanese heading "Chiakiraku" is the first natural language vs. the second natural language. In the language dictionary entry, the part of speech "noun" and bilingual expression "last performance of the seas
“On” and its part of speech “noun” are registered.

In these examples, the Japanese headline is a noun, but it does not have to be a noun, and may be a verb or an adjective. Also, in these examples, the English part-of-speech was also a noun, but this part-of-speech need not be a noun, it may be a verb or an adjective, and it need not match the part-of-speech of the headline.

Further, the first natural language-to-second natural language dictionary storage unit 10 does not have to have one first natural language-to-second natural language dictionary, but may store a plurality of dictionaries. , For example, the ALT-J / E Japanese-English dictionary and EDR
For the purpose of storing both electronic dictionaries and creating a dictionary limited to fields, for example, when creating a computer-related dictionary, it is conceivable to use a dictionary of computer terms.

Next, the function of the dictionary entry reading unit 11 will be described. The dictionary entry reading unit 11 has a function of fetching dictionary entries from one or a plurality of dictionaries stored in the first natural language versus second natural language dictionary storage unit 10 (step S2 in FIG. 14). The first natural language vs. second natural language dictionary entry read by the dictionary entry reading unit 11 is supplied to the dictionary creation management unit 12.

First natural language vs. second natural language dictionary storage unit 1
If there is only one dictionary stored in 0, for example, if Japanese is in the Japanese syllabary order, the headings are sorted and extracted in a general word order in the dictionary of the first natural language, or the first natural language vs. The two natural language dictionary entries may be divided for each part of speech of the first natural language, and the headings may be sorted and extracted within the same part of speech.

First natural language vs. second natural language dictionary storage unit 1
When a plurality of dictionaries are stored in 0, a certain search order such as the Japanese syllabary order is determined, and dictionary entries are extracted from each dictionary according to the search order, and the first natural language entry is the same in the plurality of dictionaries. If there is a first natural language to second natural language dictionary entry that is, a method of extracting the first natural language to second natural language dictionary entries from all the dictionaries can be considered.

In addition, a method may be considered in which a plurality of dictionaries are recombined into one dictionary in advance using the first natural language entry as a key, and the dictionaries are sequentially taken out as in the case of one dictionary.

In addition to the above, the stored dictionary is set to 1
There is a way to pull each one. In that case, first the first dictionary is sorted by the heading of the first natural language and drawn in order, and it is checked whether or not the first natural language entry that appears is also in another dictionary. All the first natural language-to-second natural language dictionary entries that have entries are retrieved, and the retrieved first natural language-to-second natural language dictionary entry is given a flag of retrieved. Next, after the first dictionary is searched to the end, the process moves to the next dictionary, and only the first natural language vs. second natural language dictionary entries to which the fetched flag is not added are similarly fetched and the first dictionary is removed. If the first natural language vs. second natural language dictionary entry having the first natural language entry is also registered in other dictionaries, it is taken out, and after taking out, it is taken out in each dictionary while giving the taken out flag. It is also conceivable to repeat until there are no more first natural language to second natural language dictionary entries.

Further, in particular, when the created first natural language-to-third natural language dictionary is a field-specific dictionary or when there is a difference in reliability between the dictionaries, the extracted first natural language-to-first natural language The two natural language dictionary entries are flagged as to which dictionary is included in the first natural language vs. second natural language dictionary entry, and the first natural language vs. second natural language dictionary field or reliability is used to determine the first natural language. It is also conceivable to weight the language versus the second natural language dictionary entry.

Next, the dictionary creation management unit 12 will be described. In the first natural language to second natural language dictionary entry supplied from the dictionary entry reading unit 11, the dictionary creation management unit 12 converts the second natural language entry of the first natural language to second natural language dictionary entry into the third natural language. It is replaced by the entry and stored in the first natural language-to-third natural language dictionary storage unit 13 as a set of the first natural language entry and the third natural language entry.

In the dictionary creation management unit 12, first, the entry selection information addition unit 121 refers to the first natural language corpus information storage unit 14 to refer to the first natural language versus the second natural language supplied from the dictionary entry reading unit 11. It is judged whether or not the dictionary entry is a word that frequently appears in the field of the first natural language-to-third natural language dictionary to be created, and whether it is useful to obtain as the first natural language-to-third natural language dictionary entry. If it is determined to be beneficial, the replacement execution unit 1233 is supplied.

The statistical information of the first natural language corpus stored in the first natural language corpus information storage unit 14 referred to here has a format as shown in FIG. 3, for example, and is at least in the first natural language. It shows statistical information such as a set of words and their parts of speech that appear in a certain Japanese corpus and the frequency of appearance of the set in the corpus.

If the corpus that is the source of the statistical information of the first natural language corpus stored in the first natural language corpus information storage unit 14 is Japanese, for example, it will be digitized Nihon Keizai Shimbun or Mainichi Shimbun. Can be given.

The function of the entry selection information addition unit 121 will be specifically described with reference to FIGS. 2 and 3. For example, if it is assumed that an economic dictionary is to be created, the first natural language corpus information storage unit 14 stores the statistical information acquired from the first economic natural language corpus. FIG. 3 shows an example of statistical information obtained from an economic Japanese corpus. Assuming that the entry selection information addition unit 121 is supplied with the first bank versus the second bank list of the Japanese entry “bank (noun)” or “bank employee (noun)” shown in FIG. As described above, since "bank (noun)" and "bank employee (noun)" appear in the economic-related Japanese corpus, it is judged to be a useful entry if acquired, and is supplied to the replacement executing unit 1233. (Step S in FIG. 14)
3 (Yes)).

However, if the Japanese entry of the supplied first natural language vs. second natural language dictionary entry is "burn (noun)", "four seasons (noun)" or "chiakiraku (noun)", then FIG. Since no Japanese entries appear in the economic Japanese corpus as shown in, the creation of dictionary entries corresponding to these Japanese entries is not judged to be useful when creating an economic dictionary. , The operation is moved to the next dictionary entry (step S3 (N
o)). However, if the dictionary to be created is not limited to an economic dictionary, "burn (noun)", "four seasons (noun)", and "chiakiraku (noun)" are also supplied to the replacement execution unit.

Further, the entry selection information adding section 121
By referring to the second natural language corpus information storage unit 15, the second natural language entry supplied from the dictionary entry reading unit 11 is weighted for selecting the translated word, that is, the translated word selection information is added and weighted. It also has a function of supplying the second natural language entry to the replacement executing unit 1233.

The statistical information of the second natural language corpus stored in the second natural language corpus information storage unit 15 referred to here has a format as shown in FIG. 4, for example, and at least in the second natural language. A set of words and parts of speech that appear in a certain English corpus and statistical information such as the frequency of appearance of the set in the corpus are shown.

If the corpus as the source of the statistical information of the second natural language corpus stored in the second natural language corpus information storage unit 15 is English, for example, Finantial Ti
mes (http://www.jcp.grenet.fr/ELRA/cata/text _- det.ht
ml # mlcc) and Wall Street Journal (http: //morph.ldc.up
enn.edu/Catalog/LDC2000T43.html) etc.

The weighting function of the entry selection information adding section 121 will be described in detail with reference to FIGS. 2 and 4. For example, the Japanese entry shown in Figure 2 is "burn (noun)."
"Burn (noun)" and "scale (noun)" are registered in the English entry of the first natural language versus second natural language dictionary entry. However, it can be seen from FIG. 4 that "burn (noun)" has a higher appearance frequency in the second natural language corpus. Therefore, the English translation for "burn (noun)" is "
"burn (noun)" is more important than "scale (noun)"
As a result, the French language obtained by using "burn (noun)" is emphasized rather than the French language obtained by using "scale (noun)". Therefore, for example, as shown in FIG. 5, weighting for translation word selection is performed ( Step S4 in FIG. 14).

If the statistical information of the second natural language corpus obtained from the sufficiently large second natural language corpus or the reliable second natural language corpus can be used, as in this example, the first natural language corpus can be used. It may be considered that the second natural language entry of the language-to-second natural language dictionary entry is weighted and supplied to the replacement executing unit 1233, or that only the second natural language entry having a heavy weight is supplied to the replacement executing unit 1233. To be

When another first natural language-to-second natural language dictionary entry shown in FIG. 2 is supplied to the entry selection information addition section 121 having the weighting function, the English corresponding to each Japanese entry is supplied. Since there is only one entry, the weighting function of the entry selection information addition unit 121 does not work, or when a dictionary limited to a field is created, information on the likelihood of the field of the second natural language entry is added. It is also possible to do it.

Next, the replacement execution unit 12 of the replacement management unit 123
33 will be described. The substitution executing unit 1233 uses the supplied first natural language vs. second natural language dictionary entry of the second
Using the natural language entry as a key, the second natural language vs. third natural language dictionary storage unit 19 is referred to, the third natural language entry which is a parallel translation of the second natural language entry is acquired, and the first natural language vs. second natural language is acquired. The second natural language entry of the language dictionary entry is replaced with the third natural language entry and supplied to the translated word selection information adding unit 124.

However, if the replacement executing unit 1233 cannot acquire the third natural language entry which is a parallel translation of the second natural language entry of the supplied first natural language vs. second natural language dictionary entry, the second natural language entry is replaced with the second natural language entry. The natural language entry is supplied to the second natural language analysis unit 122.

First, the second natural language-to-third natural language dictionary storage unit 19 will be specifically described with reference to FIG. 6 (a). 6A and 6B show the second natural language vs. the third.
It is an example of the second natural language to third natural language dictionary entry stored in the natural language dictionary storage unit 19.

When the second natural language is English and the third natural language is French, the second natural language-to-third natural language dictionary is, for example,
Lantmark Lexicon (LanTmark, http: //www.icp.grenet.fr
/ ELRA / cata / text _- det.html # multilex) and Basic Multili
ngual Dictionary (http://www.icp.grenet.fr/ELRA/cat
a / text _- det.html # multilex) exists.

As shown in FIGS. 6A and 6B, as the second natural language vs. third natural language dictionary entry, at least the English headline and its part of speech, and the French parallel translation expression and its part of speech for the English entry. Is registered.

For example, in FIG. 6A, the English heading is "b.
The second natural language vs. third natural language dictionary entry, which is ank, has its part of speech "noun" and its bilingual expression "banque" and its part of speech "feminine noun", and another bilingual expression "riv".
e ”and its part of speech“ female noun ”are registered. The second natural language vs. third natural language dictionary entry whose English heading is "burn" has its part of speech "noun" and bilingual expression "
ruisseau "and its part of speech" masculine noun ", and another parallel translation" bru ^ lure "and its part of speech" female noun "are registered, and the English heading is" scald ". The part of speech "noun", the parallel expression "bru ^ lure", and the part of speech "female noun" are registered in the three natural language dictionary entries.

In these examples, the English headline is a noun, but it need not necessarily be a noun, and may be a verb or an adjective. Also, in these examples, the French part of speech was also a masculine or feminine noun, but this part of speech need not be a masculine or feminine noun, and may be any verb, adjective, or even match the headline part of speech. Absent.

The second natural language-to-third natural language dictionary stored in the second natural language-to-third natural language dictionary storage unit 19 need not be one, but a plurality of dictionaries may be stored. , Such as the Lantmark Lexicon and Basic Multilingual
For the purpose of storing both dictionaries or creating a dictionary limited to fields, for example, when creating a computer-related dictionary, it is conceivable to use a dictionary of computer terms.

The replacement execution unit 1233 will be described with reference to FIG.
The function of is specifically described. Here, when the English entry "bank (noun)" for the Japanese entry "bank (noun)" shown in FIG. 2 is supplied to the replacement execution unit 1233, the second natural shown in FIG. The language-to-third natural language dictionary storage unit 19 is searched, and the French entries “banque (female noun)” and “rive” are parallel translations of “bank (noun)”.
(Female noun) "(step S5 (Y
es)), the Japanese entry “bank (noun)” and the French entries “banque (female noun)” and “rive (female noun)” are supplied to the translated word selection information addition unit 124 (step S6 in FIG. 14). ).

If the English entries "burn (noun)" and "scald (noun)" for the Japanese entry "burn (noun)" shown in FIG. 2 are supplied to the replacement execution unit 1233, the second The natural language-to-third natural language dictionary storage unit 19 is searched, and from FIG. 6A, the French entry “ruisseau (masculine noun)” for “burn (noun)” is found.

[Outer 1] And French entry "bru ^ lure" for "scald (noun)"
(Female noun) "(step S5 (Y
es)), the Japanese entry "burn (noun)" and the French entries "ruisseau (masculine noun)" and "bru ^ lure
(Female noun) ”and information indicating that“ bru ^ lure (female noun) ”appears in duplicate in the set and supplies the set to the translated word selection information adding unit 124 (step in FIG. 14). S6).

Also, "last perform" which is an English entry for the Japanese entry "Chiaki (noun)" shown in FIG.
When “ance of the season (noun)” is supplied to the replacement execution unit 1233, even if the second natural language vs. third natural language dictionary storage unit 19 is searched, the English entry is displayed as shown in FIG. 6A. There is no second to third natural language dictionary entry for and no corresponding French entry can be obtained, so this English entry "last performance of the se
The "ason (noun)" is supplied to the second natural language analysis unit 122 (step S5 (No) in FIG. 14).

Similarly, the English entry "bank clerk (noun)" for "bank employee (noun)" and the English entry "four seasons (noun)" for "four seasons (noun)"
However, since the French entry, which is a parallel translation, cannot be obtained, the English entry is supplied to the second natural language analysis unit 122 (step S5 (No) in FIG. 14).

Next, the translated word selection information adding section 124 will be described. The translated word selection information adding unit 124 verifies the third natural language corpus information storage unit 20 by using the supplied third natural language entry as a key, and determines whether the supplied third natural language entry is correct, In such a case, weighting for selecting a translated word such as which third natural language entry is prioritized is performed and stored in the first natural language to third natural language dictionary storage unit 13.

The statistical information of the third natural language corpus stored in the third natural language corpus information storage unit 20 referred to here has a format as shown in FIG. 7A, for example, and is at least the third. A set of words and parts of speech that appear in a corpus of French, which is a natural language, and statistical information such as an appearance frequency and a co-occurrence frequency in the corpus of the set are shown.

If the corpus that is the source of the statistical information of the third natural language corpus stored in the third natural language corpus information storage unit 20 is French, for example, Le Monde new
spaper (http://morph.ldc.upenn.edu/Catalog/LDC95T1
1.html) and Canadian Hansard (http: //www.ldc.upenn.
edu / ldc / news / release / hansard.html) etc.

The function of the translated word selection information adding section 124 will be specifically described with reference to FIGS. 3 and 7A. For example, if it is assumed that an economic dictionary is to be created, the statistical information acquired from the economic corpus is stored in the third natural language corpus information storage unit 20. FIG. 7A is an example of the appearance frequency acquired from the French corpus of economic relations. As described in the explanation of the entry selection information addition unit 121, when the dictionary of economic relations is created, the statistical information of the first natural language corpus of economic relations is stored in the first natural language corpus information storage unit 14. And using the statistical information,
When the first natural language vs. second natural language dictionary entries are selected, the first natural language vs. second natural language dictionary entries of the Japanese entries "bank (noun)" and "bank employee (noun)" shown in FIG. Although only remaining, the English entry corresponding to the Japanese entry “bank employee (noun)” is supplied to the second natural language analysis unit 122, and thus is supplied to the translated word selection information addition unit 124 at this stage. There is only a Japanese entry "bank (noun)" and a French entry obtained from "bank (noun)" which is the only English entry for the Japanese entry.

Therefore, the Japanese entry "bank (noun)"
And French entries “banque” and “rive”
If the set of "(feminine nouns)" is supplied to the translated word selection information addition unit 124, "banque (feminine nouns)" appears frequently in the economic-related French corpus from FIG. 7A.
It turns out that "rive (female noun)" does not appear, and the French entry for "bank (noun)" is "banque
(Female noun) ”is more promising.
The English "bank (noun)" has a meaning such as "shore" in addition to the meaning of "bank", and "rive (female noun)" is a French entry that appears in the latter meaning. In this way, the French entries "banque (female noun)" and "rive (female noun)" are weighted for selecting the translated words, and the Japanese entry "bank (noun)" and the first natural language vs. The data is stored in the 3 natural language dictionary storage unit 13 (step S7 in FIG. 14).

If the statistical information of the third natural language corpus obtained from a sufficiently large third natural language corpus or a reliable third natural language corpus can be used, the weight of the third natural language entry is heavy. It is also conceivable that only the first natural language-to-third natural language dictionary storage unit 13 is stored, leaving only the ones.

However, in the above example, the statistical information of the first natural language corpus is stored in the first natural language corpus information storage unit 14, and the first natural language versus the second natural language is used by using the statistical information. Assuming that dictionary entries have already been selected, the first natural language corpus information storage unit 1
4 does not exist, or even if the first natural language corpus information storage unit 14 exists and the first natural language vs. second natural language dictionary entry is not selected, the third natural language corpus information storage unit similarly. By using statistical information such as the appearance frequency of the third natural language corpus stored in 20, the French language entries can be selected or weighted.

Also, the translated word selection information adding section 124
Even if there are a plurality of second natural language entries of the natural language vs. second natural language dictionary entry, if the supplied third natural language entry appears in duplicate, the third appearing in duplicate
It has a duplication judgment function that judges that natural language is more appropriate.

The duplication determination function of the translated word selection information adding section 124 will be specifically described. As described above, the English entry for the Japanese entry "burn (noun)" shown in FIG. 2 "
From "scald (noun)", the French entry "bru ^ lure (female noun)" shown in FIG.
From "(noun)", the French entry "ruis" shown in FIG.
seau (masculine noun) "and" bru ^ lure (feminine noun) "
Is acquired and supplied to the translated word selection information addition unit 124. In the supplied French language entry, only "bru ^ lure (feminine noun)" appears in duplicate. Therefore, the French entry for "burn (noun)" is "ruisseau
We weight "bru ^ lure (feminine noun)" as more promising than "masculine noun". In fact, in English "burn (noun)", it means "brook" in addition to "burn". "Ruisseau (masculine noun)" is a French word meaning Ogawa. In this way, the French entries "bru ^ lure (feminine noun)" and "ruisseau (masculine noun)" are weighted for selecting the translated words, and the Japanese entry "yaku (noun)" and the first natural language It is stored in the third natural language dictionary storage unit 13 (step S7 in FIG. 14). However, it is possible to store only the third natural language entry having a heavy weight in the first natural language to third natural language dictionary storage unit 13.

Next, the second natural language analysis section 122 will be described. In the second natural language analysis unit 122, it is supplied when the replacement management unit 123 cannot replace the second natural language entry of the first natural language vs. second natural language dictionary entry with the third natural language entry. Second
The natural language entry is analyzed, and the analysis result of the second natural language entry is supplied to the replacement management unit 123 again.

Examples of the analysis method of the second natural language analysis section 122 include morphological analysis and syntactic analysis. As a morphological analyzer for English, Brill Tagg
er (Eric Brill, ftp: blaze.cs.jhu.edu: / pub / brill /)
Or Xelda (Xerox, ftp: //ftp.parc.xerox.com/pub/tagge
r) etc., and ApplePi for parsing English.
e Parser (New York University, http: //cs.nyu.edu/cs/
projects / proteus / app /) and Link Grammar (Carnegie M
ellon University, http://www.link.cs.cmu.edu/link
/) And so on.

First, regarding the case where the second natural language analysis unit 122 performs the morphological analysis and the syntax analysis, FIGS.
The function of the second natural language analysis unit 122 will be specifically described using. 8A to 8C show the second natural language analysis unit 12
6 is an example of the analysis result of the second natural language entry according to No. 2.

The English entry "last performance of th" for the Japanese entry "Chiaki Raku (noun)" shown in FIG.
It is assumed that "e season (noun)" is supplied to the second natural language analysis unit 122. FIG. 8 (a) shows "last performance of".
The morphological analysis result of the second natural language analysis unit 122 of "the season" is shown. Further, Fig. 8B shows the syntactic analysis result of the second natural language analysis unit 122 of the English entry. According to the morphological analysis result shown in (a), "last performance of the season" is "last
(Adjective) ”,“ performance (noun) ”,“ of (preposition) ”,“ the (definite article) ”,“ season (noun) ”are subjected to morphological analysis, and from FIG. ) "+" Season (noun) "becomes a noun phrase, and" of "
(Preposition) ”is added to form a prepositional phrase, and also“ last
(Adjective) "+" performance (noun) "becomes a noun phrase, and" last (adjective) performance (noun) of (preposition) the (definite article) season (noun) "becomes a noun phrase. The result is supplied to the replacement management unit 123 (step S10 in FIG. 14).

As another analysis method of the second natural language analysis unit 122, for example, it is conceivable that only the morpheme analysis is performed and each morpheme is passed to the substitution management unit 123. In this case, for example, the English entry "bank clerk (noun)" for the Japanese entry "bank employee (noun)" shown in FIG.
Is a "bank (noun)", "cler" as shown in Fig. 8 (c).
Morphological analysis is performed on "k (noun)" and the result is supplied to the replacement management unit 123 (step S10 in FIG. 14).

Further, as an analysis method of the second natural language analysis section 122, for example, in the case of a compound word, it is possible to divide each word constituting the compound word and pass it to the substitution management section 123. In that case, the English entry “bank clerk” for the Japanese entry “bank employee (noun)” shown in FIG.
"(Noun)" is divided into "bank" and "clerk" and replaced
It is supplied to the management unit 123 (step S1 in FIG. 14).
0).

Further, as the analysis method of the second natural language analysis unit 122, for example, in the case of a compound word, it is possible to divide it into subsets and pass them to the substitution management unit 123. In that case, the Japanese entry "Chiakiraku (noun)" shown in Figure 2
English entry for "last performance of the seaso
n (noun) ”,“ last performance of the ”,“ la
st performance of ”,” last performance ”,” las
t ”,“ performance of the season ”,“ parformance
of the ”,“ performance of ”,“ performance ”,“ o
f the season ”,” of the ”,” of ”,” the seaso
It is divided into n ”,“ the ”, and“ season ”, and the replacement management unit 1
23 (step S10 in FIG. 14).

In addition, the second natural language analysis section 122 refers to the second natural language dictionary storage section 17 for storing the dictionary of the second natural language in FIG. It has a second natural language dictionary utilization analysis function capable of analyzing a natural language entry.

The second natural language dictionary stored in the second natural language dictionary storage unit 17 stores at least the original form, part of speech, and utilization information of the second natural language entry. Figure 9
Is an example of an English dictionary entry stored in the second natural language dictionary storage unit 17, and shows information on the original form of English, the part of speech, and its variation.

The second natural language analysis unit 122 will be described with reference to FIG.
The above-described second natural language dictionary utilization analysis function of is specifically described. When the English entry "four seasons (noun)" for the Japanese entry "four seasons (noun)" shown in FIG. 2 is supplied, the second natural language analysis unit 122 divides the supplied English entry into word units, From the second natural language dictionary storage unit 17 shown in FIG. 9, the information that “four” is an adjective invariant and “seasons” is a plural form of “season (noun)” is obtained, and the substitution management unit 123 is obtained. Supply to (Fig. 1
4 step S10).

Next, the replacement management unit 123 will be described. In FIG. 1, as already described in the description of the replacement execution unit 1233,
When the second natural language entry of the second natural language entry is supplied to the replacement management unit 123 without passing through the second natural language analysis unit 122, the second natural language entry is supplied to the replacement execution unit 1233 and replaced. The executing unit 1233 stores the second natural language vs. second natural language dictionary stored in the third natural language dictionary storage unit 19.
Searching a natural language vs. a third natural language dictionary to obtain a third natural language entry corresponding to the second natural language entry,
The second natural language entry of the first natural language vs. second natural language dictionary entry is replaced with the third natural language entry, and the set of the first natural language entry and the third natural language entry is supplied to the translation word selection information addition unit 124. To do.

When the replacement management unit 123 is supplied with the analysis result of the second natural language entry, the analysis result is supplied to the replacement rule searching unit 1231. The replacement rule search unit 1231 searches the replacement rule storage unit 16 using the supplied analysis result of the second natural language entry as a key, and obtains the third natural language entry from the analysis result of the second natural language entry. Try to get more than one. When the replacement rule is acquired, the analysis result of the second natural language entry and the replacement rule are supplied to the replacement execution unit 1233, and the replacement execution unit 1233 receives the second natural language of the first natural language versus the second natural language dictionary entry. Entry is third
The natural language entry is replaced, and the combination of the first natural language entry and the third natural language entry is added to the translation word selection information adding unit 1.
Supply to 24.

In the replacement rule creating unit 1232, the replacement rule searching unit 1231 acquires a plurality of replacement rules,
When the replacement rules are used in combination, the number of appearances of the combination of the replacement rules is counted, and when the combination of the replacement rules (combination replacement rule) appears a certain number of times or more, the combination is stored in the replacement rule storage unit 16. .

The replacement rules stored in the replacement rule storage unit 16 are manually created in advance or machine learning using a decision tree or the like (for example, "Programs for Machine") is performed.
Learning ”, Morgan Kaufman, 1992) or the one created by machine learning is manually modified and stored. Further, as described above, the same combination replacement rule is generated by the replacement rule search unit 1231. When the combination replacement rule appears more than a certain number of times, the combination replacement rule is also stored in the replacement rule storage unit 16. It may be possible to manually check the combination replacement rule.

The function of the replacement management unit 123 will be described in detail with reference to FIGS. 8B, 10 and 6B. It is assumed that the replacement management unit 123 is supplied with the English entry "last performance of the season (noun)" for "Chiaki (noun)". FIG. 8B shows the analysis result supplied from the second natural language analysis unit 122.
0 is an example of the replacement rule stored in the replacement rule storage unit 16, and FIG. 6B shows the dictionary entry of the English-French dictionary stored in the second natural language vs. third natural language dictionary storage unit 19. Here is an example.

In the replacement rule search unit 1231, for example,
A method of giving priority to the replacement rule that can be searched in the longest unit can be considered. In this case, "last" shown in FIG.
From the result of parsing "performance of the season",
First, the replacement rule of "last (adjective) performance (noun) of (preposition) the (definite article) season (noun)" is searched, but it is not stored in the replacement rule storage unit 16 as shown in FIG. The next longest unit, "of
Search for a replacement rule of "(preposition) the (definite article) season" (noun). For this, the condition 1 of the replacement rules of rule numbers 5 and 6 is matched.
“Season (noun)” is searched from the second natural language vs. third natural language dictionary storage unit 19 shown in FIG. 6B, and the French entry “saison” corresponding to the English entry “season (noun)”.
(Female noun) ”is acquired. Since the part of speech of the French entry is“ female noun ”, the replacement rule 6 is met. So "of (preposition) the (definitive article) season (noun)" is "d
e la saison ”(French prepositional phrase)”.

Next, similarly, the remaining "last (adjective) perf
ormance (noun) "replacement rule. For this, condition 1 of the replacement rules of rule numbers 1 and 2 matches, and the second natural language vs. third natural language dictionary shown in FIG. "Repre'sentation (female noun)" and "spectacle (masculine noun)" are acquired from the storage unit 19 as French entries corresponding to "performance (noun)".
So "last (adjective) performance (noun)" is "d
ernie′re repre′sentation ”or“ dernie′r spectacle ”.

"De la saison" (French preposition phrase) and "dernie're repre'sentation" obtained through the above substitutions
(French noun phrase) ”or“ dernie′r spectacle ”(French noun phrase)” is searched for a replacement rule, and “dernie′re repre′sentat” is used using the replacement rule of rule number 7.
"ion de la saison" or "dernie'r
spectacle de la saison ”can be obtained (step S12 (Yes) in FIG. 14).

From the above, "dernie're repr" as the French entry for the Japanese entry "Chiakiraku (noun)"
e′sentatione de la saison (noun phrase) ”or“ d
ernie'r spectacle de la saison ”.

However, when the second natural language vs. third natural language dictionary storage unit 19 stores a plurality of dictionaries, "pe
As a French entry corresponding to "rformance (noun)",
For example, "repre'sentation (feminine noun)" is listed in any dictionary, but "spectacle (masculine noun)" is listed only in some dictionaries.
"tion (feminine noun)" is given priority, and eventually "dernie're"
It may be possible to give priority to “repre′sentation dela saison”.

In addition, the "last performance of the se"
In the replacement for “ason”, among the replacement rules shown in FIG. 10, rule number 6 and rule number 1 or 2,
And because the rule number 7 is used in combination,
The combination replacement rule shown in FIG. 11 can be created. If this combination replacement rule appears more than, for example, a predetermined number of times (step S14 (Y
es)), and stored in the replacement rule storage unit 16 (FIG. 14).
Step S15).

Assuming that a dictionary of economic relations is created, if the morphological analysis result of "bank clerk" shown in FIG. 8C is supplied to the substitution management unit 123, "
It can be replaced in the same manner as in the case of "last performance of the season" (using the replacement rules of rule numbers 8 and 9 shown in FIG. 10). As a result, the French entry for the Japanese entry "bank employee (noun)" is " em
ploye ′ de banque ”and“ employe′e de banque ”
And “greffier de banque” and “employe′de riv
e ”and“ employe′e de rive ”and“ greffier de
You can get rive ”.

Further, the replacement management unit 123 searches the second natural language-to-third natural language dictionary storage unit 19 using the supplied analysis result of the second natural language entry as a key, and further searches the third natural language dictionary storage unit 19.
It has a third natural language dictionary search / replace function for searching the natural language dictionary storage unit 18 and replacing the second natural language entry with the third natural language entry.

The third natural language dictionary stored in the third natural language dictionary storage unit 18 stores at least the original form, part of speech, and utilization information of the third natural language entry. Figure 1
Reference numeral 2 is an example of a French dictionary entry stored in the third natural language dictionary storage unit 18, and stores information on the original form, part of speech, and variation forms of the French language.

The replacement management unit 123 having the above function will be described in detail with reference to FIG. The English entry “fo” for the Japanese entry “Shiki (noun)” shown in FIG.
The analysis result for "ur seasons (noun)" is the replacement management unit 1.
Consider the case where it is supplied to 23. As the analysis result, as described in the explanation of the second natural language analysis unit 122 having the second natural language dictionary utilization analysis function, "four seasons" is obtained.
"(Noun)" is divided into "four (adjective)" and "seasons (noun)", and "four" is an adjective unchanged from the second natural language dictionary storage unit 17 shown in FIG.
It is assumed that information that "s" is a plural form of "season (noun)" is supplied (step S10 in FIG. 14).

First, the second natural language vs. third natural language dictionary storage unit 19 shown in FIG. 6 (b) is searched, and the English entry "fo
Acquire the French entries "quatre" and "saison" for "ur" (adjective) and "season" (noun), where "season" is originally "se"
"asons" and plural forms, so "saison (feminine noun)"
Search for plurals in Figure 12 to get "saisons" and "quatr
e saisons ”.

However, in the above example, in the replacement management unit 123 having the third natural language dictionary search / replacement function, the second natural language analysis unit 122 is stored in the second natural language dictionary storage unit 17. The replacement was performed using the analysis result of the second natural language entry performed using the natural language dictionary. However, the analysis result of the second natural language entry supplied to this replacement management unit 123 is not necessarily the second natural language. The analysis result using the dictionary storage unit 17 does not have to be used, and may be an analysis result obtained by another analysis method of the second natural language analysis unit 122, such as a result of morphological analysis.

In addition, using only the heading of the second natural language entry as a key, the second natural language vs. third natural language dictionary storage unit 19 is searched, and the third natural language entry that matches only in the character face is extracted. Depending on the analysis method used by the second natural language analysis unit 122, various substitution methods can be considered.

In this way, when the second natural language entry of the first natural language to second natural language dictionary entry can be replaced with the third natural language entry, the first natural language entry and the third natural language entry are replaced. The set is supplied to the translated word selection information addition unit 124.

However, in the replacement execution unit 1233, the first
Even if the second natural language entry of the natural language vs. second natural language dictionary entry could not be replaced with the third natural language entry, the analysis method that has not been performed for the second natural language entry is the second natural language analysis. If it remains in the unit 122, the second natural language entry is returned to the second natural language analysis unit 122, and the same processing is performed again (step S9 (Yes) in FIG. 14).

Next, the translated word selection information adding section 124 will be additionally described. As described above, the translated word selection information adding unit 124 searches the third natural language corpus information storage unit 20 using the supplied third natural language entry as a key, and stores the stored statistical information of the third natural language corpus. Is used to determine the likelihood of the acquired third natural language entry, or when a plurality of third natural language entries are supplied, weighting for translation selection is performed, and the acquired first natural language entry is acquired. And a set of third natural language entries is first
Store in the natural language vs. third natural language dictionary storage unit 13.

As described above, in the example of the Japanese entry "bank (noun)" shown in FIG. 2, the French entry obtained from the English entry "bank (noun)" corresponding to the Japanese entry is the third natural entry. The case where the weighting is performed using the appearance frequency in the French corpus as shown in FIG. 7A stored in the language corpus information storage unit 20 has been described.

Here, an example in which the co-occurrence frequency in the third natural language corpus as shown in FIG. 7B is used as the statistical information of the third natural language corpus will be described. For example, if it is assumed that an economic dictionary is to be created, the third natural language corpus information storage unit 20 stores the statistical information acquired from the third economic natural corpus. FIG. 7B is an example of the co-occurrence frequency acquired from the French corpus of economic relations, and at least French language 1 and its part of speech 1, French language 2 and its part of speech 2 and their co-occurrence frequency are stored. .

The function of the translated word selection information adding section 124 will be specifically described with reference to FIG. 7B. Suppose you want to create an economic dictionary. Here, in the translation word selection information adding unit 124, "employe 'de banque" and "employe'e de ban" are given as French entries for "bank employee (noun)".
que ”and“ greffier de banque ”and“ employe ′ d
e rive ”and“ employe′e de rive ”and“ greffi
er de rive ”is supplied. When the third natural language corpus information storage unit 20 is searched, as shown in FIG. 7B, in the French corpus of economic relations,“ employe '(masculine noun) ”and“ banque ”are provided. (Feminine noun), "and" employ
"e'e (feminine)" and "hanque (feminine)" co-occur, but not others. Therefore, among the French entries supplied, "employe 'de banque" and "e"
It can be judged that "mploye'e de banque" is promising (step S7 in FIG. 14).

Regarding "rive (feminine noun)",
As described in the above example of “bank (noun)”, FIG.
It can also be seen from (a) that the French entry using "banque (female noun)" is more promising (step S7 in FIG. 14).

The same weighting is applied to the example of "Chiaki (noun)". However, the French corpus used in this case is not limited to economic relations. Regarding the "four seasons (noun)" example, only one French language entry was acquired, but in order to verify the likelihood of the acquired French language entry, it may be possible to refer to the French language corpus for weighting as well. (Step S7 in FIG. 14).

These results are stored in the first natural language-to-third natural language dictionary storage unit 13.

Finally, the first natural language-to-third natural language dictionary storage unit 13 will be described. The first natural language-to-third natural language dictionary storage unit 13 needs at least the headline of the first natural language and its part of speech, the parallel expression of the third natural language corresponding to the first natural language entry, and its part of speech. In this case, the translation selection information of the third natural language can also be stored.

FIGS. 13A and 13B are examples of the first natural language to third natural language dictionary entry stored in the first natural language to third natural language dictionary storage unit 13. FIG.
(A) is the first natural language Japanese heading and its part of speech,
In this example, the bilingual expression and its part of speech for the Japanese entry of French, which is the third natural language, are stored. FIG.
In (a), only promising French entries are stored in the first natural language to third natural language dictionary storage unit 13 (step S8 in FIG. 14). FIG. 13B is an example in which non-promising French language entries are also stored with weighting information for translation word selection. However, if you want to leave only promising French entries, or if you also want to leave unpromising French entries, you can set a certain threshold in advance and cut off the French entries whose weights for selecting translated words are below the threshold. Conceivable. Also in FIG. 13B, “greffier de banque” and “emplo” acquired as French entries for the Japanese entry “bank employee (noun)”.
y′e de rive ”and“ employe′e de rive ”and“
“Greffier de rive” is rounded down because the weight is below the threshold. It is also possible to manually check the created 1st natural language vs. 3rd natural language dictionary.

The above-described processing is carried out until there are no more dictionary entries stored in the first natural language versus second natural language dictionary storage unit 10, and when there are no more dictionary entries, the processing is terminated (step S1).

The processing procedure of the dictionary creating method of the above embodiment is recorded as a program in a recording medium, the recording medium is incorporated into a computer system, and the program recorded in the recording medium is downloaded or installed in the computer system. However, by operating the computer system with the program, it is possible to function as a dictionary creating device that implements the dictionary creating method, and by using such a recording medium, the distribution of the dictionary can be improved. It is possible.

[0151]

As described above, according to the present invention,
A first natural language-to-second natural language dictionary entry is retrieved from the first natural language-to-second natural language dictionary storage unit and corresponds to the second natural language entry of the retrieved first natural language-to-second natural language dictionary entry. Second natural language entry to second
Natural language vs. third natural language dictionary storage, obtained by replacing the second natural language entry of the first natural language vs. second natural language dictionary entry with this third natural language entry, and the headline of the first natural language and its part of speech And a first natural language to third natural language bilingual dictionary having a third natural language bilingual translation which is a bilingual translation of the first natural language entry and its part-of-speech, and creating a first natural language to third natural language dictionary. The first natural language to the third natural language dictionary and the second natural language to the third natural language dictionary are used to store the first natural language to the third natural language dictionary.
The first natural language-to-third natural language dictionary can be created without a dictionary or bilingual corpus of combinations of natural languages.

Further, according to the present invention, the first natural language corpus information storage unit is referred to by using the first natural language entry of the first natural language to second natural language dictionary entry supplied from the dictionary entry reading unit as a key. , The unnecessary first natural language vs. second natural language entry is removed, so that, for example, when creating a domain-specific dictionary, unnecessary dictionary entries are not created, and as a result, the first natural language is deleted. It is possible to speed up and improve efficiency in creating and using the third natural language dictionary. Further, when a dictionary limited to a field is created, if the first natural language vs. second natural language dictionary and the second natural language vs. third natural language dictionary to be used can also use the dictionary of the field, it is more reliable. Field limited first
A natural language to a third natural language dictionary can be created.

Further, according to the present invention, the second natural language corpus information storage unit is referred to by using the second natural language entry of the first natural language vs. second natural language dictionary entry supplied from the dictionary entry reading step as a key. Since the translation word selection information, which is the weighting information, is added, even when there are a plurality of translation word candidates from the first natural language to the second natural language, which translation word is more appropriate as, for example, a technical term or a field-specific word. It is possible to know which one is present, and to prioritize which translation word candidate appears more frequently among the plurality of translation word candidates. Further, even when there is only one translation word candidate, the likelihood of the translation word candidate can be determined, and as a result, a more reliable first natural language versus third natural language dictionary can be created. In addition, since information for selecting a translated word is added at the stage of the second natural language, the same information can be used when changing only the third natural language without changing the first natural language and the second natural language. As a result, when the third natural language is expanded into various languages, the first natural language-to-third natural language dictionary can be efficiently created.

According to the present invention, if the second natural language entry of the first natural language to second natural language dictionary entry cannot be replaced by the third natural language entry, then the second natural language entry is analyzed. , The third natural language entry is acquired based on this analysis result, and the acquired third natural language entry replaces the second natural language entry.
For example, even if the second natural language entry is a compound word and is not registered in the second natural language vs. third natural language dictionary as it is, it is used as a partial expression unit forming a compound word for the second natural language entry. If each partial expression is registered in the second natural language-to-third natural language dictionary by performing analysis such as division, the second natural language entry of the first natural language-to-second natural language dictionary entry is changed to the third natural language entry. It can be replaced by a language entry, and as a result, a third natural language entry for a compound word or the like that is difficult to replace by the notation as it is can also be acquired to create a first natural language to third natural language dictionary.

Further, according to the present invention, the replacement rule storage unit is searched based on the analysis result of the second natural language analysis unit, the replacement rule is extracted, and the extracted replacement rule is used to make a plurality of replacement rules. In this case, the third natural language entry is acquired from the second natural language vs. third natural language dictionary storage unit by using the plurality of substitution rules in combination, so that, for example, the second natural language entry is a compound word or a variation. Alternatively, even if the combination of them is not registered as it is in the second natural language to third natural language dictionary, the analysis result of the second natural language entry, the replacement rule suitable for the analysis result, and the second natural language. The third natural language dictionary is used to ensure more reliable first even for more complicated second natural language entries.
The second natural language entry of the natural language to second natural language dictionary entry may be replaced with the third natural language entry. In addition, even if the replacement rule that can replace the analysis result of the second natural language entry in a lump by using a plurality of replacement rules in combination is not stored in the replacement rule storage unit, the first natural language vs. second natural language A third natural language entry for a more complex second natural language entry, such as a compound word, which can replace the second natural language entry of the natural language dictionary entry with the third natural language entry, and as a result is difficult to replace as it is. The first natural language versus the third natural language dictionary can be created accurately. Furthermore, because you can combine and use replacement rules,
It is not necessary to store all possible replacement rules in the replacement rule storage unit, and the first natural language vs. third natural language dictionary can be efficiently created.

Further, according to the present invention, when a plurality of replacement rules are used in combination, the number of times of use of the combined replacement rules is counted, and when the number of uses exceeds a predetermined number, the combined replacement is performed. Since the rule is stored in the replacement rule storage unit, a combination replacement rule that is frequently used is acquired and stored in the replacement rule storage unit, so that the second natural language of the first natural language versus the second natural language dictionary entry is stored. It is possible to increase the speed at which the entry is replaced with the third natural language entry.
A natural language dictionary can be efficiently created.

According to the present invention, it has a second natural language dictionary which stores the original form, part of speech, and utilization information of the second natural language, and the second natural language analysis section searches the second natural language dictionary storage section. Then, since the second natural language entry is analyzed, for example, even if the second natural language entry includes a variation and is not registered in the second natural language vs. third natural language dictionary as it is, the second natural language entry is analyzed. When the original form of the variation included in the second natural language entry can be obtained from the natural language dictionary and the original form is registered in the second natural language vs. third natural language dictionary, the first natural language vs. the second natural language. The second natural language entry of the dictionary entry can be replaced with the third natural language entry, and as a result, the third natural language entry corresponding to the second natural language entry such as a variation that is difficult to replace in its original form can be obtained. Can, it is possible to create precisely the first natural language versus the third natural language dictionary.

Further, according to the present invention, it has a third natural language dictionary storing the original form, part of speech, and utilization information of the third natural language, and the analysis result of the second natural language analysis unit or the replacement rule search. Since the third natural language dictionary storage unit is searched based on the replacement rule extracted by the section to replace the second natural language entry of the first natural language vs. second natural language dictionary entry with the third natural language entry, for example, the second natural language entry When the natural language entry includes a variation and is not registered in the second natural language vs. third natural language dictionary as it is, the variation included in the second natural language entry is corrected to the original form,
When the corresponding third natural language entry is obtained in the original form from the second natural language-to-third natural language dictionary, the third natural language dictionary is searched to change the original form of the third natural language entry into an appropriate variation notation. As a result, the third natural language entry corresponding to the second natural language entry such as a variation that is difficult to replace as it is can be converted into an accurate expression, and the first natural language to the third natural language dictionary can be converted. Can be created accurately.

Furthermore, according to the present invention, a plurality of second natural language entries of the first natural language vs. second natural language dictionary entry are obtained, and the third natural language entry for the second natural language entry is duplicated. If it is acquired, it is determined that the redundantly acquired third natural language entry is more appropriate. Therefore, when there are a plurality of translation word candidates at the stage of the second natural language entry, a plurality of second word candidates are acquired. If the third natural language entry acquired from the natural language entry is duplicated, translation word selection information that the duplicated third natural language entry is more appropriate than the third natural language entry that appears alone can be added, As a result, a highly reliable first natural language-to-third natural language dictionary can be created.

According to the present invention, there is provided a third natural language corpus information storage section for storing statistical information of the third natural language corpus, and the information of the third natural language corpus information storage section is used to Since the translated word selection information indicating whether or not the third natural language entry acquired by the dictionary creation management unit is appropriate is added, during the tracing from the first natural language entry to the third natural language entry via the second natural language entry. Even if the number of translation candidates increases, it is possible to know which translation word is more appropriate, such as a technical term or a field-specific word. In addition, it is possible to give priority to a word that is used more frequently among a plurality of translation word candidates, and further, even if there is only one translation word candidate in the third natural language, the likelihood of the translation word candidate can be determined. Furthermore, when the second natural language entry is a compound word and the third natural language entry cannot be acquired by the notation as it is, and the analysis is performed to acquire the third natural language entry, word candidates in the third natural language entry Even if there are a plurality of words, a more appropriate word combination can be selected from the co-occurrence information between words included in the acquired third natural language entry, resulting in a highly reliable first natural language vs. 3 A natural language dictionary can be created.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a dictionary creation device according to an embodiment of the present invention.

FIG. 2 is a diagram showing first natural language to second natural language dictionary entries stored in a first natural language to second natural language dictionary storage unit used in the dictionary creating apparatus of FIG.

FIG. 3 is a diagram showing first natural language corpus information stored in a first natural language corpus information storage unit used in the dictionary creation device of FIG.

FIG. 4 is a diagram showing second natural language corpus information stored in a second natural language corpus information storage unit used in the dictionary creation device of FIG.

5 is a diagram showing an execution result of an entry selection information addition unit used in the dictionary creation device of FIG.

6 is a diagram showing a second natural language to third natural language dictionary entry stored in a second natural language to third natural language dictionary storage unit used in the dictionary creation device of FIG. 1;

7 is a diagram showing third natural language corpus information stored in a third natural language corpus information storage unit used in the dictionary creation device of FIG. 1. FIG.

8 is a diagram showing an analysis result of a second natural language entry by a second natural language analysis unit used in the dictionary creation device of FIG.

9 is a diagram showing a second natural language dictionary entry stored in a second natural language dictionary used in the dictionary creation device of FIG. 1. FIG.

FIG. 10 is a diagram showing replacement rules stored in a replacement rule storage unit used in the dictionary creation device of FIG. 1.

FIG. 11 is a diagram showing a replacement rule used in a replacement rule search unit used in the dictionary creation device of FIG. 1.

FIG. 12 is a diagram showing a third natural language dictionary entry stored in a third natural language dictionary storage unit used in the dictionary creation device of FIG. 1.

13 is a diagram showing first natural language-to-third natural language dictionary entries stored in a first natural language-to-third natural language dictionary storage unit used in the dictionary creating apparatus of FIG.

FIG. 14 is a flowchart showing an operation of the dictionary creation device of FIG.

[Explanation of symbols]

10 first natural language vs. second natural language dictionary storage 11 Dictionary entry reading section 12 Dictionary creation management unit 13 First Natural Language Versus Third Natural Language Dictionary Storage 14 First Natural Language Corpus Information Storage 15 Second Natural Language Corpus Information Storage 16 Replacement rule storage 17 Second Natural Language Dictionary Storage 18 Third Natural Language Dictionary Storage 19 second natural language vs. third natural language dictionary storage 20 Third Natural Language Corpus Information Storage 121 entry selection information addition unit 122 Second Natural Language Analysis Unit 123 Replacement Management Department 124 Translation word selection information addition unit 1231 Replacement Rule Search Unit 1232 Replacement rule creation unit 1233 replacement execution unit

─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-2-301868 (JP, A) Kumiko Tanaka, Kyoji Umemura, Hideya Iwasaki, Creation of bilingual dictionary via third language, Journal of Information Processing Society, Japan, June 15, 1998, Vol. 39, No. 6, p. 1915-p. 1924 (58) Fields surveyed (Int.Cl. ⁷ , DB name) G06F 17/21-17/28 JISST file (JOIS)

Claims (13)

(57) [Claims] 1. A dictionary creating apparatus for creating a bilingual dictionary of natural language pairs, comprising a headline of a first natural language and its part of speech, a bilingual expression of a second natural language which is a bilingual translation of the first natural language entry, and A first natural language to second natural language dictionary storage for storing a dictionary of the first natural language to the second natural language having the part of speech, a heading of the second natural language and its part of speech, and the second natural language entry A second natural language pair third natural language dictionary storing a third natural language dictionary having a third natural language bilingual expression and its part-of-speech which are parallel translations of the first natural language pair; A dictionary entry reading unit that extracts a first natural language-to-second natural language dictionary entry from a second natural language dictionary storage unit and a second natural language entry of the extracted first natural language-to-second natural language dictionary entry. Obtaining a third natural language entry from the second natural language to third natural language dictionary storage and replacing the second natural language entry of the first natural language to second natural language dictionary entry with the third natural language entry. A dictionary creation management unit and a first natural language to third natural language dictionary entry generated by replacing the second natural language entry of the first natural language entry to second natural language dictionary entry with the third natural language entry. Based on the first natural language headline and its part-of-speech, and the third natural language parallel translation and its part-of-speech which are parallel translations of the first natural language entry, a first natural language vs. third natural language bilingual dictionary is created. A first natural language-to-third natural language dictionary storage unit for storing, and a first natural language corpus information storage unit for storing statistical information of the first natural language corpus; The creation management unit refers to the first natural language corpus information storage unit with the first natural language entry of the first natural language to second natural language dictionary entry supplied from the dictionary entry reading unit as a key, and unnecessary A dictionary creating apparatus having an entry selection information addition unit for removing a first natural language vs. a second natural language entry. 2. A second natural language corpus information storage unit for storing statistical information of a second natural language corpus, wherein the entry selection information addition unit is provided with the first natural language corpus information supplied from the dictionary entry reading unit. The second natural language entry of the natural language pair second natural language dictionary entry as a key
2. The dictionary creating apparatus according to claim 1, further comprising: translated word selection information / weighted information adding means for referring to the natural language corpus information storage unit and adding translated word selection information which is weighting information. 3. The second natural language entry when the second natural language entry of the first natural language vs. second natural language dictionary entry cannot be replaced by the third natural language entry. A second natural language analysis unit for analyzing the third and a third natural language analysis unit based on the analysis result of the second natural language analysis unit.
3. The dictionary creating apparatus according to claim 1, further comprising a replacement execution unit that acquires a natural language entry and replaces the second natural language entry with the acquired third natural language entry. 4. A replacement rule storage unit storing a replacement rule for replacing the second natural language with the third natural language, wherein the dictionary creation management unit stores the analysis result of the second natural language analysis unit. A replacement rule search unit that searches the replacement rule storage unit based on the replacement rule storage unit and extracts a replacement rule; and the replacement execution unit uses the replacement rule extracted by the replacement rule search unit, and when there are a plurality of replacement rules 4. The dictionary creation according to claim 3, further comprising means for acquiring a third natural language entry from the second natural language vs. third natural language dictionary storage unit by using the plurality of substitution rules in combination. apparatus. 5. When the dictionary creation management unit uses a plurality of replacement rules retrieved by the replacement rule search unit in the replacement execution unit in combination, the usage count is used to count the number of times of use of the combined replacement rules. Means and
The dictionary according to claim 4, further comprising: a replacement rule creating unit that stores the combined replacement rule in the replacement rule storage unit when the counted number of times of use of the replacement rule exceeds a predetermined number. Creation device. 6. A second natural language dictionary storage unit for storing a second natural language dictionary storing an original form, a part of speech, and utilization information of the second natural language, wherein the second natural language analysis unit is provided. 6. The dictionary creating apparatus according to claim 3, further comprising means for searching the second natural language dictionary storage unit and analyzing the second natural language entry. 7. A third natural language dictionary storage unit for storing a third natural language dictionary storing an original form, a part of speech, and inflection information of the third natural language, wherein the replacement execution unit is the second A second natural language entry of the first natural language versus second natural language dictionary entry is searched by searching the third natural language dictionary storage unit based on the analysis result of the natural language analysis unit or the replacement rule retrieved by the replacement rule searching unit. 7. The dictionary creating apparatus according to claim 3, further comprising a unit that replaces with a third natural language entry. 8. The dictionary creation management unit obtains a plurality of second natural language entries of the first natural language vs. second natural language dictionary entry, and sets a third natural language entry for the second natural language entry.
When the natural language entry is acquired redundantly, the translation word selection information addition unit for determining that the third natural language entry acquired redundantly is more appropriate is provided. 7. The dictionary creation device according to any one of 7. 9. A third natural language corpus information storage unit that stores statistical information of a third natural language corpus, wherein the dictionary creation management unit stores the information of the third natural language corpus information storage unit. 3. A means for adding translated word selection information indicating whether or not the third natural language entry acquired by the dictionary creation management unit is appropriate by using the dictionary.
9. The dictionary creation device according to any one of 8 to 8. 10. A dictionary creation method for creating a bilingual dictionary of natural language pairs, comprising a headline of a first natural language and its part of speech, and a bilingual expression of a second natural language which is a bilingual translation of the first natural language entry, and A first natural language vs. second natural language dictionary storage section storing a first natural language vs. second natural language dictionary having the part of speech in a first natural language vs. second natural language dictionary storage section; A second natural language vs. third natural language dictionary having a natural language heading and its part of speech, and a third natural language bilingual expression which is a parallel translation of the second natural language entry and its part of speech is used as the second natural language vs. third. Second stored in the natural language dictionary storage unit
A step of storing a natural language vs. a third natural language dictionary storage section; a step of reading a first natural language vs. a second natural language dictionary entry from the first natural language vs. second natural language dictionary storage section; A third natural language entry corresponding to the second natural language entry of the first natural language vs. second natural language dictionary entry is obtained from the second natural language vs. third natural language dictionary storage, and the first natural language vs. A dictionary creation management step of replacing the second natural language entry of the second natural language dictionary entry with the third natural language entry, and the second natural language entry of the first natural language vs. second natural language dictionary entry to the third natural language entry And the part of speech of the first natural language based on the first natural language-to-third natural language dictionary entry generated by replacing A first natural language to third natural language dictionary storage step of creating and storing a first natural language to third natural language bilingual dictionary having a third natural language bilingual translation which is a bilingual translation of a natural language entry and its part of speech; Storing the statistical information of the first natural language corpus in the first natural language corpus information storage unit, the dictionary creation management step including the first natural language pair second supplied from the dictionary entry reading step. An entry selection information providing step of removing unnecessary first natural language vs. second natural language entries by referring to the first natural language corpus information storage unit using the first natural language entry of the natural language dictionary entry as a key. Characteristic dictionary creation method. 11. The method further comprises a step of storing statistical information of a second natural language corpus in a second natural language corpus information storage section, wherein the entry selection information assigning step is the first supplied from the dictionary entry reading step. Natural language vs. second
A translation word selection information / weighting information assigning step of assigning translation word selection information which is weighting information by referring to the second natural language corpus information storage unit using the second natural language entry of the natural language dictionary entry as a key; The dictionary creating method according to claim 10, characterized in that 12. The dictionary creation management step, if the second natural language entry of the first natural language to second natural language dictionary entry cannot be replaced with the third natural language entry, the second natural language entry. And a second natural language analysis step for analyzing the second natural language analysis step, and a third natural language entry is acquired based on the analysis result of the second natural language analysis step, and the second natural language entry is replaced by the acquired third natural language entry. 12. The dictionary creating method according to claim 10, further comprising an executing step. 13. A recording medium in which a dictionary creating program for creating a bilingual dictionary of natural language pairs is recorded, wherein a headline of the first natural language and its part of speech, and a second natural translation which is a bilingual translation of the first natural language entry. Storing a first natural language to second natural language dictionary having a bilingual expression of a language and its part of speech in a first natural language to second natural language dictionary storage section, first natural language to second natural language dictionary storage section A second natural language-to-third natural language dictionary having a step, a headline of the second natural language and its part of speech, a parallel translation of a third natural language which is a parallel translation of the second natural language entry, and its part of speech; Natural language vs. second natural language dictionary stored in third natural language dictionary storage unit
A step of storing a natural language vs. a third natural language dictionary storage section; a step of reading a first natural language vs. a second natural language dictionary entry from the first natural language vs. second natural language dictionary storage section; A third natural language entry corresponding to the second natural language entry of the first natural language vs. second natural language dictionary entry is obtained from the second natural language vs. third natural language dictionary storage, and the first natural language vs. A dictionary creation management step of replacing the second natural language entry of the second natural language dictionary entry with the third natural language entry, and the second natural language entry of the first natural language vs. second natural language dictionary entry to the third natural language entry And the part of speech of the first natural language based on the first natural language-to-third natural language dictionary entry generated by replacing A first natural language to third natural language dictionary storage step of creating and storing a first natural language to third natural language bilingual dictionary having a third natural language bilingual translation which is a bilingual translation of a natural language entry and its part of speech; Storing the statistical information of the first natural language corpus in the first natural language corpus information storage unit, the dictionary creation management step including the first natural language pair second supplied from the dictionary entry reading step. An entry selection information providing step of removing unnecessary first natural language vs. second natural language entries by referring to the first natural language corpus information storage unit using the first natural language entry of the natural language dictionary entry as a key. A recording medium in which a characteristic dictionary creation program is recorded.