JP2003122748A - Automatic translation pre-processing system, pre- processing method, program for pre-processing or recording medium in which the same program is recorded - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a pre-processing system, a pre-processing method, a program for the pre-processing for contriving efficient translation by performing automatic pre-processing before applying a Japanese document having terms and codes to inhibit machine translation to the machine translation and for further realizing enhancement of quality or a recording medium in which the same program is recorded. SOLUTION: The pre-processing such as (1) converting terms to inhibit the machine translation, terms having unnecessary phrasing and redundant terms, etc., into brief terms and converting KANA (Japanese syllabary) terms and abbreviated terms into normal KANJI (Chinese character) terms, (2) inserting translation unnecessity control codes into unit codes, codes and Greek characters, etc., (3) inserting control codes for forced sentence division into titles, etc., and complicated long sentences, etc., is performed to the Japanese document having terms, codes and complicated long sentences to inhibit the machine translation by using the automatic pre-processing system. Enhancement of efficiency of a translation work and enhancement of quality of the translation are achieved by applying the pre-processed document to general purpose machine translation.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an automatic translation preprocessing system, a preprocessing method and a computer for a Japanese document having terms, symbols and complicated long sentences that inhibit translation by a programmed computer. The present invention relates to a program for preprocessing or a recording medium recording the program.

[0002]

2. Description of the Related Art Currently, commercially available translation software for personal computers can perform automatic sentence segmentation processing, but it is hardly useful for improving the quality of translations obtained. The preprocessing of sentences to be translated is performed manually by a person, which requires a great deal of labor and skill, which is a big obstacle especially when processing a large amount of documents. Terms that hinder machine translation,
If a Japanese document having symbols or long sentences is omitted and pre-processed and machine-translated as it is, the quality of the obtained translated sentence is remarkably low, and a great deal of labor is required for post-processing such as correction and correction.

[0003]

SUMMARY OF THE INVENTION The object of the present invention is to achieve a great deal by pre-processing a Japanese document having terms, symbols and complicated long sentences that hinder machine translation by a computer, prior to machine translation. Does not require great effort and skill,
It is an object of the present invention to provide a preprocessing system, a preprocessing method, a program for preprocessing, or a recording medium recording the program, which can improve the quality of the obtained translated text.

[0004]

SUMMARY OF THE INVENTION The present invention is visual
A program for preprocessing is constructed using a basic (VB) tool, and Japanese documents are processed through this program prior to machine translation.

Specifically, it is composed of the following (1) to (3). (1) A system for preprocessing a Japanese document having terms, symbols, and complicated long sentences that obstruct machine translation by a programmed computer, and displays an input document and an output document Display means and database that accumulates terms, symbols and long sentences that hinder machine translation,
Data that accumulates easily translated terms corresponding to the inhibitory term
A table, a first search means for searching a term that inhibits machine translation in an input document based on a database that inhibits machine translation, and a database that accumulates symbols and long sentences; When an inhibitory term is retrieved, a means for substituting the term into an easily translatable term and a symbol in the input document are retrieved based on the accumulated database of the terms easily translated corresponding to the inhibitory term. Second search means for searching and a control code for a translation-unnecessary process at a corresponding position of the symbol when the symbol is searched.
Means for automatically inserting a code, a third searching means for searching a long sentence in an input document, and a control code for forcibly dividing the sentence at the corresponding portion when the long sentence is searched. An automatic translation preprocessing system comprising means for inserting and means for outputting a document for which preprocessing has been completed.

(2) A database that accumulates terms that inhibit machine translation, symbols and long sentences, and a database that accumulates easily translatable terms corresponding to the inhibiting terms are constructed.
A method for preprocessing a Japanese document having a term, a symbol or a long sentence, which inhibits machine translation by a programmed computer, comprising a display step of displaying an input document and an output document, and a machine. A step of retrieving a term that inhibits machine translation in an input document based on a term that inhibits translation, a database that accumulates symbols and long sentences,
When the inhibitory term is searched, a step of replacing the easily translatable term corresponding to the inhibitory term with an easily translatable term, and then a symbol in the input document And a second search step of searching for the symbol, a step of automatically inserting a control code for the translation-unnecessary process at the relevant portion of the symbol when the symbol is searched, and then searching for a long sentence in the input document. And a step of automatically inserting a control code for forcibly dividing the sentence at the relevant position when a long sentence is retrieved, and a step of outputting the document for which preprocessing is completed. Automatic translation preprocessing method.

(3) A computer database constructed by accumulating terms that inhibit machine translation, symbols and long sentences, and accumulating easily translatable terms corresponding to the inhibiting terms. Is a program for causing a computer to pre-process a Japanese document having a term, a symbol or a long sentence that inhibits machine translation, wherein the program inhibits the machine translation. Based on the database that accumulates terms, symbols and long sentences, search for a term that inhibits machine translation in the input document, and if the inhibiting term is found, translate it corresponding to the inhibiting term. Based on the accumulated database of easy terms, replace them with easy-to-translate terms, then search for the symbol in the input document, and if the symbol is found, translate it to the corresponding part of the symbol. Control of unnecessary processing Automatically injects the, then, to find the long sentence in the input document, if the long sentence has been retrieved, control to enforce division processing sentences relevant sections co -
A program for preprocessing a Japanese document, or a recording medium having the program recorded therein, characterized in that a document is automatically inserted and a preprocessed document is output.

[0008]

BEST MODE FOR CARRYING OUT THE INVENTION Terms that inhibit translation, terms such as symbols and long sentences are extracted from a large amount of test data by repeating a machine translation test of Japanese documents, and those with a high statistical probability are listed. Then, the data is converted into a database, and the automatic conversion preprocessing is performed based on this. Of course, in parallel with this, it is necessary to store the term registration in the user dictionary, and the use of both will achieve the object of the present invention.

The present invention will be described below based on specific examples. <System Configuration> As the computer to be used, a general-purpose personal computer may be used. The preprocessing program is a Visual Basic V.6.0 (Professional Edition
option). This program causes a computer to search for a term that inhibits machine translation in an input document on the basis of a database that accumulates symbols and long sentences that inhibit machine translation, and Is searched for, the translatable term corresponding to the inhibitory term is replaced with the translatable term based on the accumulated database, and then the symbol in the input document is searched for. When a symbol is searched, a control code for the translation-free process is automatically inserted at the corresponding position of the symbol, and then a long sentence in the input document is searched. When a long sentence is searched, the corresponding position is searched. The control code for forcibly dividing the sentence is automatically inserted, and the document for which the preprocessing is completed is output. Here, terms that inhibit machine translation, databases that accumulate symbols and long sentences, and databases that accumulate easy-to-translate terms corresponding to the inhibiting terms are constructed in advance in a computer. deep.

In the present invention, the term that inhibits machine translation includes the term for translation inhibition exemplified in Table 1 below, unnecessary terminology exemplified in Table 2, redundant term and terminology peculiar to patent specification. Means a unit symbol, a numbered symbol, Greek letters, equal signs used in mathematical formulas, inequality sign addition, subtraction, multiplication and division symbols, chemical structural formulas, chemical reaction formulas, etc. that do not need to be translated. . In addition, the long sentence means a complicated long sentence such as a title of a patent specification or the like. Further, since the control code differs depending on the translation software used, it is necessary to change it according to the translation software used. In the present invention, the translation software used in the specific examples is Toshiba The Translation Professional V.6. 6.0 was used.

<Document Example> As a Japanese document having terms, symbols, and long sentences that hinder translation, the abstract of the published patent publication was used. This requires a large amount of processing, and there are a large number of translation-inhibiting terms, unnecessary terminology, redundant terms, and patent-specific terminology, symbols, etc. Can't get Patent Gazette CD
-The one that downloaded the abstract from ROM etc. was used.

<Pre-Processing> The abstract sentence of the Japanese Patent Laid-Open Publication is read into a programmed computer, and a database for accumulating words, symbols and complicated long sentences which inhibit the translation accumulated in the computer are read. Based on the text, the sentence is searched, and when a disturbing term is found, a process of automatically replacing the term with a preset term is performed. Also, when a symbol is searched, the process is performed. A process that automatically inserts a control code that does not require translation (hereinafter referred to as a translation-unnecessary processing code) to the relevant part of the symbol, and when a complex long sentence is searched, the relevant part of the sentence A computer is caused to automatically insert a control code for forcibly dividing a sentence (hereinafter referred to as a forced sentence dividing process code). The control code includes the above-mentioned translation unnecessary processing code and the forced sentence division processing code, and the forced sentence division processing code also includes the parallel phrase designation code, the insertion phrase designation code and the like. This compulsory text division processing code inserts a control code for designating a text division location and a specific style. The purpose of compulsory sentence division is to reduce confusion of phrases in sentences as much as possible during machine translation,
The processing is performed based on a preset condition.

[0013] The standard styles of patent publication abstracts issued so far are:

[Problem] and

It consists of, but in addition to this

[Problem] and

SOLUTION: and

【effect】,

[Solution] and

[Effect] These are not unified in many types, such as those with full text in front of them, and those with no sentence division. Also the expression

【Purpose】

There is also a structure. this house

[Issue] A summary of missing sentences

[Effect] sentence

PROBLEM TO BE SOLVED: To insert a compulsory sentence division processing code so that a sentence is replaced and translated.

The English version corresponding to the abstract of the patent publication is PAJ (Pa
tent abstract of Japan), but this has a unified style, TITLE (name of invention), PURPOSE (issue),
It is composed of CONSTITUTION. Also, PU
The RPOSE sentence is unified into a style that begins with the infinitive To + (the original form of the verb). At the same time, the process for designating the style is performed by the compulsory sentence division process code.

The translated sentence that has been machine-translated is post-processed by a usual method. However, if the above-mentioned summary of the patent publication has a special format, it needs to be reworked. Items that are not classified,
Items that have some paragraphs but are lacking in sentences, and those that only describe the conventional problems, etc., require significant reworking and cannot be solved unless the patent gazette abstract is improved. However, for most of the abstracts of patent publications excluding these, the translation work time can be significantly improved by performing the preprocessing of the present invention.

[0016] Examples of terms that inhibit machine translation are given in Table 1 below. For example, "contains" or "contains" means "contains" when subjected to machine translation.
Alternatively, it is translated as two verbs "contained" and "naru", which disturbs the whole translated sentence. Of course, it is only necessary to register the corresponding translated words for each word in the user dictionary, but it is very inefficient to register all the large numbers of terms.
If you can replace only the ending of each term, for example, "do" is changed to "do", and "do" is "do"
If replaced with, the efficiency of the pretreatment work is greatly improved.

<Examples of Unnecessary Wording Terms> Examples are given in Tables 2 and 3 described later. For example, "a method for manufacturing a composite material which is a method of infiltrating a molten metal" is "The manufa
cturing method of the composite material it was pr
esupposed that it is the method of infiltrating mo
It becomes a lten metal "and becomes a meaningless translation. Replacing this with "consisting of" with "consisting of" means "the manufacturing method of the composite mate"
rial which composed the method of infiltrating
You can get a simple translation like "a molten metal". In addition, some wording peculiar to a patent specification is necessary for a full-text translation of a patent that requires strictness, but in the abstract sentence, the sentence is complicated, so it is easier to replace it with simple terms Then, it becomes easy to understand. For example, "it can be configured", "is intended", "proposes", "reduces can be reduced", "occurs" Will disappear ”etc.

<Unit symbol etc.> For example, the translation of "volume%" does not make sense because "volume" and "%" are separated from each other. This can be solved by the replacement process and the translation unnecessary code insertion process.
Units containing numbers adversely affect the overall translation. m2, m
The translation of m2 etc. is m [two or more] or m ... ・ by 2. etc. These can be solved by inserting a translation unnecessary processing code in the preprocessing step. In addition, if you want to leave the element symbol as a symbol instead of spelling, you can solve it by inserting a translation unnecessary processing code.

<Blank between symbols> For example, the translation of “JIS G 0561” is far from JIS. G ... by 0561, which adversely affects the overall translation. This can also be solved by performing a blank search in the preprocessing stage and, if detected, performing a blank automatic removal process.

<Sentence-stopped sentence> Some of the problem sentences and solution sentences have a sentence stop at the end, but such a sentence-stopped sentence does not translate well. In this case, it can be solved by supplementing appropriate terms in preprocessing. For example, “providing.” Becomes “providing.” And “method.”
To “provide a method”, “device.” To “provide a device”, and the like.

<Forced Text Division> As mentioned above, the patent publication
The abstract for translation output from the CD-ROM is the patent application publication number, the title of the invention, and the abstract (consisting of the problem, the solution, etc.). However, other items can be added depending on the purpose of translation. When this output summary sentence is directly subjected to machine translation, the automatic sentence segmentation function of commercially available translation software is not sufficient, and many "term parts" and sentences are connected without being segmented. When translated as it is, this “term part” is taken into the “subject” and “object” of the translated sentence, and the translated sentence is often confused. Therefore, this problem can be solved by inserting a control code for forcible division in advance in advance.

For example, (1) "

SOLUTION: The superalloy welding composition of CONSTITU is ...
TION: the invention is ... this

[Solution] After "." Code is inserted automatically and compulsory sentence division, the translation is CONSTITUTION: The superalloy
welding composition of the invention is ...

(2) "

PROBLEM TO BE SOLVED: Generally, in the main part of a relatively large structure in the field of civil engineering or construction ...
or the main sections of the comparatively large si
zed structure in engineering works or a constructi
on fieldat the general PURPOSE:. Similarly,

[Problem] After "." Code is automatically inserted and compulsory sentence division, the translation is PURPOSE: Generally ... for the ma
in sections of the comparatively large sized struc
ture in engineering works or a construction field.
Becomes

(3) "

A high-strength steel sheet having a plate thickness of 20 mm or less and high impact absorption energy. Translation for "High-strength sheet steel w
hich has high impact-absorption energy by 20mm or
less of PURPOSE: plate thicknesses. Similarly,

[Problem] After "." Code is automatically inserted to compulsorily divide sentences, and at the same time to correct sentence-stop sentences, "steel plate"
Automatically inserts the term "provide" after
Furthermore, in order to use the PAJ method (To + infinitive) for the style of translation
* The code and the parallel phrase designating code "{}" are automatically inserted at the specified position designated by the text search.

【Task】. * Provide "{High-strength steel plate with a thickness of 20 mm or less and high impact absorption energy}". ". This translation is `` To provide the high-strength sheet steel
which has high impact-absorption energy in 20mm or
less of plate thicknesses. "

(4) In order to capitalize the title of the invention, the translation of "metal-ceramic composite material manufacturing method" is "Ma
nufacturing method of a metal / ceramic composite ma
"*" code is automatically inserted at the beginning of the sentence, the translation is "MANUFACTURING METHOD OF METAL / CERAMIC C
OMPOSITE MATERIAL ”.

[0026]

[Table 1] Examples of terms that inhibit machine translation (Example of translation inhibition term) (Example of replacement term) You can do it With or without Can be done Will be done Was done Done Do as you can Lesser less less Do it like

[0027]

[Table 2] Examples of unnecessary phrases (Example of unnecessary wording) (Example of replacement term) To let Let me let you let me To do To do Will do To do Was decided to be Try to be

[0028]

[Table 3] Examples of wording specific to patents (Example of wording unique to patent) (Example of replacement term) Can be configured to Get to get The challenge is to achieve Is intended to Make possible make possible Equipped with Is intended to provide Can be reduced, not reduced Give it to give

[0029]

EFFECTS OF THE INVENTION In machine translation of a large amount of sentences such as the abstract of a patent publication, it is necessary to correct in advance the factors of terms and symbols that hinder translation. By using the automatic translation preprocessing system, the automatic translation preprocessing method or the program for automatic translation preprocessing of the present invention,
This work efficiency and work time are greatly improved, and at the same time, the quality of machine translation is improved, so the work required for post-processing can be greatly reduced.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an automatic translation preprocessing system.

Claims (3)

[Claims] 1. A system for preprocessing a Japanese document having a term, a symbol, or a complex long sentence that impedes machine translation by a programmed computer, which processes an input document and an output document. Display means for displaying, database for accumulating terms, symbols and long sentences that inhibit machine translation, database for accumulating easy-to-translate terms corresponding to the inhibiting term, and inhibiting machine translation First searching means for searching for a term that inhibits machine translation in an input document based on a database that accumulates a term, a symbol or a long sentence, and if the inhibiting term is found, A replacing means for replacing the easily translatable term corresponding to the obstructing term into the easily translatable term, and a second searching means for retrieving the symbol in the inputted document; , The symbol was searched In this case, means for automatically inserting a translation unnecessary processing control code at the relevant portion of the symbol, and third searching means for searching for a long sentence in the input document,
When a long sentence is searched, a means for automatically inserting a compulsory sentence segmentation control code for forcibly segmenting the sentence at the corresponding position and a means for outputting the document for which preprocessing has been completed are provided. Automatic translation preprocessing system. 2. A database for accumulating terms that inhibit machine translation, symbols and long sentences, and a database for accumulating easily translatable terms corresponding to the inhibition terms have been constructed and programmed. A method of preprocessing a Japanese document having a term, a symbol or a long sentence, which inhibits machine translation, by a computer, a display step of displaying an input document and an output document, and a machine translation inhibiting process. A step of searching for a term that inhibits machine translation in an input document based on a database that accumulates terms, symbols and long sentences, and, if the inhibiting term is searched, corresponds to the inhibiting term A step of substituting a term that is easy to translate into a term that is easy to translate based on the accumulated database, a step of searching a symbol in the input document, and a step of searching the symbol when the symbol is searched, Where the symbol applies A step of automatically inserting a translation-free processing control code, then a third search step of searching for a long sentence in the input document, and a forced division processing of the sentence at the corresponding portion when a long sentence is searched. An automatic translation preprocessing method comprising: a step of automatically inserting the control code for forced sentence segmentation processing, and a step of outputting a document for which preprocessing has been completed. 3. A computer in which a database for accumulating terms that inhibit machine translation, symbols and long sentences and a database for accumulating easy-to-translate terms corresponding to the inhibiting terms are constructed. A program for causing a computer to pre-process a Japanese document having a term that inhibits machine translation, a symbol or a long sentence, the program being a term that inhibits machine translation, Based on a database that accumulates symbols and long sentences, search for a term that inhibits machine translation in an input document, and when the inhibiting term is searched, a term that is easy to translate corresponding to the inhibiting term Is replaced with a term that is easy to translate based on the accumulated database, and then the symbol in the input document is searched, and when the symbol is searched, the translation-free processing is performed at the corresponding position of the symbol. Control code of When a long sentence is searched for, the forced sentence segmentation control code for forcibly segmenting the sentence at the corresponding position is automatically inserted. A program for preprocessing a Japanese document or a recording medium recording the program, which is characterized in that the preprocessed document is output.