JP2006065542A - Machine translation method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a machine translation method suitable for the translation of many languages for quickening translation processing, for acquiring a natural text as a translation result, and for simplifying customization for each user. <P>SOLUTION: This machine translation is provided to divide an inputted original language sentence into a noun removal sentence and a noun, and to retrieve a condition-added corresponding type database and a noun database for translating it into the noun removal sentence and the noun of an object language, and for translating the original language sentence into the object language sentence only by combining the noun removal sentence of the object language with the noun of the object language. All words and phrases other than a noun are treated as a noun removal sentence being a simple character string, and a syntax analysis or a meaning analysis is not operated. As for the condition-added corresponding type database and noun database, they are only retrieved, respectively, and any analysis not operated. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a machine translation method for translating a source language into a target language, and more particularly to a machine translation method suitable for translation between multiple languages.

The modern society is an advanced information society, and with the development of communication technology including the Internet, it has become possible to obtain information from around the world instantly without geographical restrictions. There are many cases that are written in other languages, and the demand for machine translation is increasing. Various services have already been put into practical use for machine translation, and most are rule-based translation methods based on grammatical rules (see Japanese Patent Application Laid-Open No. 2004-86919). However, in the rule-based translation method, it is difficult to say that the user is satisfied because the translation processing takes time and an unnatural translation is created.
JP 2004-86919 A

Machine translation users want machine translation methods that are easy to operate, have a fast translation process, get natural translations, are easily customized according to their usage, and are suitable for multilingual translation. is there. However, since the conventional rule-based machine translation method performs processing based on syntax analysis and semantic analysis, it takes time for the translation processing, and the translation result has a limit to meet the user's request. On the other hand, machine translation methods that do not require syntactic analysis and semantic analysis have been invented (see Japanese Patent Application Laid-Open Nos. 2002-7392, 2002-7393, and 2002-7395). However, these methods also analyze verbs including tense and determine imperative sentences / question sentences based on sentence patterns, so the problem remains that the translation process cannot be simplified.
JP 2002-7392 A JP 2002-7393 A JP 2002-7395 A

  In view of the above-described situation, the object of the present invention is to provide a machine that is capable of high-speed translation processing, produces natural sentences as translation results, is easily customized for each user, and is suitable for translation between multiple languages. To provide a translation method.

  The inventor has continued intensive research for many years in order to solve the above-mentioned problems, and has developed a machine translation method and system that does not require syntactic analysis and semantic analysis, leading to the following invention.

A first invention is a machine translation method for translating a source language into a target language,
(1) creating a conditional correspondence expression database for storing a conditional correspondence expression that associates a noun removal sentence in a source language with a noun removal sentence in a target language;
(2) creating a noun database that stores source language nouns, target language nouns corresponding to the source language nouns, and nouns corresponding to the source language nouns and nouns corresponding to the target language nouns; ,
(3) inputting a source language sentence as a translation target;
(4) extracting a noun from the source language sentence and dividing it into a source language noun removal sentence and a source language noun;
(5) searching a noun database based on the source language noun to determine attributes of the source language noun, and translating the source language noun into a target language noun;
(6) Based on the noun removal sentence in the source language, search the conditional correspondence database using the noun attribute as a condition, determine the conditional correspondence expression, and use the noun removal sentence in the source language as the target language Translating into noun-removed sentences,
(7) introducing the target language noun obtained in step (5) into the target language noun removal sentence, and creating a target language sentence;
(8) a step of outputting the target language sentence as a translation result;
Is a machine translation method consisting of

  In the second invention, the conditional correspondence database is a collection of a plurality of conditional correspondence databases classified by style, and the user selects one or more conditional correspondence databases according to the purpose of use. The translation method described in the first invention is characterized in that it is used as a translation method.

  A third invention is the machine translation method according to any one of the first to second inventions, wherein a user can add a conditional correspondence expression to the conditional correspondence expression database.

  A fourth invention is characterized in that the noun database is an aggregate of a plurality of noun databases classified according to a field, and the user selects and uses one or more noun databases according to the purpose of use. The machine translation method according to any one of the first to third inventions.

  A fifth invention is the machine translation method described in any one of the first to fourth inventions, wherein a user can add a noun to the noun database.

  The sixth invention registers the items in the conditional correspondence database and the noun database in phonological order, and searches the conditional correspondence database and the noun database in order from the front of the character string. The machine translation method according to any one of the first to fifth inventions, wherein a phoneme is determined for each character.

A seventh invention is a machine translation system that translates a source language into a target language,
(1) an input device for inputting a source language sentence to be translated;
(2) an output device that outputs a target language sentence as a translation result;
(3) a conditional correspondence expression database for storing a conditional correspondence expression that associates a noun removal sentence in the source language with a noun removal sentence in the target language;
(4) a noun database storing source language nouns, target language nouns corresponding to the source language nouns, attributes of the source language nouns and nouns corresponding to the target language nouns;
(5) A translation engine that searches the conditional correspondence database and the noun database based on the input source language sentence and performs translation processing into the target language.
Is a machine translation system consisting of

  The eighth invention is characterized in that the conditional correspondence database, the noun database, and the translation engine are stored in a server and accessed from each communication terminal and used. It is a machine translation system described in the invention.

  A ninth invention is the machine translation system according to the eighth invention, wherein the communication terminal is a mobile phone.

  Prior to the description of the present invention, the terms “noun removal sentence”, “attribute”, and “conditional correspondence expression” are defined.

  The “noun removal sentence” in the present invention indicates a remaining part obtained by removing a noun from a source language sentence or a target language sentence. All “noun removal sentences” composed of words including verbs and particles other than nouns are treated as simple character strings.

  The “attribute” of a noun in the present invention is classified by the concept of a noun such as a person, an animal, a place, a tool, food, and transportation. When one noun can be a plurality of “attributes”, it has a priority according to the appearance frequency.

  The “conditional correspondence expression” in the present invention is an expression that associates a noun removal sentence in the source language with a noun removal sentence in the target language, and is a sentence in which noun attributes are added to the part where the noun originally exists. is there.

  “Conditional expressions” that use the same verb are treated as different “conditional expressions” when the word order or words are different, such as question sentences, imperative sentences, or changes in tense, but the sentence endings are slightly different. However, if the contents are the same (for example, go, go, etc.), they can be saved as a single set in the conditional correspondence database.

Machine translation based on the present invention divides the input source language sentence into noun removal sentences and nouns, and translates into noun removal sentences and nouns of the target language by searching the conditional correspondence database and noun database, The translation into the target language sentence is performed only by combining the noun removal sentence of the target language and the noun of the target language.
All phrases other than nouns are treated as noun removal sentences that are simply character strings, and syntax analysis and semantic analysis are not performed. For conditional correspondence expressions and nouns, only the respective databases are searched, and no analysis is performed.

Since the conventional rule-based machine translation method is indispensable, translation can be performed without the need for syntactic analysis and semantic analysis, so that a high processing speed can be obtained.

The conditional correspondence expressions stored in the conditional correspondence expression database are originally entered by dividing the natural sentences of each language into noun removal sentences and attributes one by one. can get.

  A conditional correspondence formula can be easily constructed by inputting a bilingual sentence, and a conditional correspondence formula database and a noun database can be added. Therefore, customization for each user is easy.

  Machine translation based on the present invention uses only a conditional correspondence database and a noun database that are input by dividing a natural sentence of each language into a noun removal sentence and an attribute, and it is not necessary to consider grammar. Suitable for multilingual translation.

  A block diagram of a machine translation system according to the present invention is shown in FIG.

The machine translation system shown in FIG.
An input device 100 for inputting a source language sentence to be translated;
An output device 200 for outputting a target language sentence as a translation result;
A translation engine 300 that searches a conditional correspondence database and a noun database based on the input source language sentence, and performs a translation process into a target language;
A conditional correspondence expression database 400 for storing a conditional correspondence expression that associates a noun removal sentence in the source language with a noun removal sentence in the target language;
A noun database 500 that stores source language nouns, target language nouns corresponding to the source language nouns, and nouns of the source language nouns and nouns corresponding to the target language nouns;
Is provided.

  The input device 100 may be anything that is normally used for computer operations, such as a keyboard, an external storage medium, and download from the Internet. Furthermore, a voice input system may be used in combination.

  The output device 200 may be any device such as a display or a printer that is normally used for operating a computer. Furthermore, an audio output system may be used in combination.

The conditional correspondence database 400 may be a collection 410 of sentence-specific conditional correspondence databases classified by style.
In the machine translation according to the present invention, it is desirable that the conditional correspondence database is enriched in order to create a natural translation. However, when using machine translation at universities and research institutes, it is often the case that academic papers and lecture manuscripts are targeted, and literary expressions are important. On the other hand, the Internet and mobile phones are used on a daily basis. When using machine translation, colloquial expressions are expected to be used more frequently.
Conditional correspondence expressions are stored separately for each style, such as conversation style and paper style, and each database can be selected by selecting the style-specific conditional correspondence database according to the purpose of use when performing machine translation. Can be reduced, and the processing speed can be further improved.

  Furthermore, a user-specific conditional expression database 420 to which a user can add conditional correspondence expressions is set, and processing is performed in the same manner as the sentence-specific conditional correspondence expression database. It is easy to customize according to your needs.

The noun database 500 may be a collection 510 of field-specific noun databases classified by field.
The number of vocabularies in the noun database is a major factor that affects the translation ability, but it can be said that the required phrases are completely different when translating academic papers and translating everyday conversations. In particular, to translate academic papers, it is desirable that vocabulary is enriched in each specialized field.
Nouns are divided and stored by fields such as biochemistry and mechanical engineering, and by making it possible to select the field-specific noun database according to the purpose of use when translating, the capacity of each database is reduced and processed. The speed can be further improved.

  Furthermore, by setting a user-specific noun database 520 to which a user can add nouns and performing the same processing as the field-specific noun database, customization according to the needs of each user is facilitated.

Each item in the conditional correspondence database and the noun database is registered in phonological order, and when searching the conditional correspondence database and the noun database, phonemes are determined for each character in order from the front of the character string. Therefore, the only factor that affects the search time is the number of characters in the target noun removal sentence or noun, and a high-speed search is possible regardless of the number of items in the database. .
When searching the conditional correspondence database based on the noun removal sentence, the character string before and after the noun is divided and searched.

The machine translation system for carrying out the present invention is not only used in a single computer but also in an environment in which a common server can be accessed from a communication terminal used by an individual, such as on an intranet or the Internet. The correspondence type database, the noun database, and the translation engine may be stored in a server, and the server may be accessed and used from a communication terminal used by each individual. By storing the conditional correspondence database that requires storage capacity and the noun database in the server, the burden on each communication terminal is reduced.

  Furthermore, the communication terminal may be a mobile terminal such as a mobile phone or a PDA. The machine translation according to the present invention is a machine translation method that is easy for translation processing, has a small memory capacity and power consumption, and is suitable for use in a portable terminal such as a mobile phone or a PDA.

The process of translating between multiple languages will be described by taking as an example the case where the input sentence is Japanese and translated into English and Chinese.
The notation used in the sentence means J: Japanese sentence, E: English sentence, C: Chinese sentence, j: Japanese noun, e: English noun, c: Chinese noun, x: noun (attribute), respectively. Yes.
J (x), E (x), and C (x) are conditional correspondence equations. In the conditional correspondence formula, nouns have only meanings, so they are expressed using x as nouns that do not consider language.

  A machine translation flowchart is shown in FIG. The machine translation process is as described in claim 1. It is assumed that the creation of the conditional correspondence database and the noun database described in the steps (1) and (2) has been completed. Table 1 shows the conditional correspondence database, and Table 2 shows the noun database.

Step (3) A source language sentence is input as a translation target.
As an example, suppose that a Japanese sentence “He took a taxi to the station” was entered.

Step (4) A noun is extracted from the source language sentence, and is divided into a source language noun removal sentence and a source language noun.
The noun removal sentence in the source language is “J: j1 rides on j3 until j2”, and the extracted noun is “j1 = hi, j2 = station, j3 = taxi”.

Step (5) Based on the nouns in the source language, the noun database is searched to determine the attributes of the nouns in the source language, and the nouns in the source language are translated into nouns in the target language.
The noun database is searched, and the attributes of each noun are “x1: person, x2: place, x3: means of transportation”. When translated into English and Chinese nouns, “e1 = he, e2 = the station, e3 = a” taxi ”,“ c1 = other, c2 = vehicle, c3 = departure car ”.

Step (6) Based on the noun removal sentence in the source language, search the conditional correspondence database using the noun attribute as a condition, determine the conditional correspondence expression, and aim for the noun removal sentence in the source language Translate to language noun removal sentences.
Based on the noun removal sentence “J: j1 rides j3 up to j2”, the conditional correspondence database is searched using the attributes “x1: person, x2: location, x3: transportation” for each noun. Then, the conditional correspondence formula is “J (x) = x1 is on x3 up to x2. (Condition x1: Unconditional, x2: Location, x3: Transportation means)”.
Furthermore, when translating Japanese noun removal sentences into English and Chinese noun removal sentences,
The results of “E: e1 talk e3 to e2.” And “C: c1 raised to c3 termination c2.” Are obtained.

Step (7) The target language noun obtained in step (5) is introduced into the target language noun removal sentence to create a target language sentence.
The English noun removal sentence “E: e1 talk e3 to e2.” Introduces the English noun “e1 = he, e2 = the station, e3 = a taxi” to create an English sentence.
The Chinese noun removal sentence “C: c1 raised c3 finished c2.” Introduces the Chinese noun “c1 = other, c2 = vehicle, c3 = departure car” to create a Chinese sentence.

Step (8) The target language sentence is output as a translation result.
As the translation result for “J: He took a taxi to the station” entered as the translation target, the English sentence “E: He talk a taxi to the station.”
The Chinese sentence “C: other departure taxi exit vehicle” is displayed.

  Japanese particles and English prepositions have the same function, but the words do not correspond one-to-one. Therefore, conventional machine translation does not provide sufficient accuracy, and countermeasures are desired. In the present invention, particles and prepositions are distinguished by attributes set in the noun database.

Japanese sentence a) “J: ~ eat at the cafeteria” and b) “J: ~ eat with a knife” are translated into English as an example to explain the distinction of prepositions. The translation result is shown first.
B) “E: ~ eat“ in ”the dining room”,
B) “E: ~ eat“ with ”a knife ~”
It is.

Both noun removal sentences in Japanese are “J: ~ j1” “eat”, so if you do not consider the attributes of nouns, noun removal sentences in the target language,
B) “E: ~ eat“ in ”e1”,
B) “E: ~ eat“ with ”e1”
Cannot be distinguished. (See Table 1)

Here, by setting an attribute item in the noun database,
B) “J1 = Dining room” has “Place”
B) “Tool” for “j1 = knife”
The attribute is added.

Even if the noun removal sentence is the same, the conditional correspondence formula is different depending on the condition, that is, the noun attribute,
A) “Eat at ~ x1 ~ (x1: place) = ~ eat“ in ”x1-”,
B) "Eat with x1 ~ (x1: tool) = ~ eat" with "x1""
And can be distinguished from each other.

A database search processing method will be described. In the present invention, when the items in the conditional correspondence database and the noun database are registered in the phonemic order and searched, the phoneme is determined for each character in order from the front of the character string.
When searching the conditional correspondence database based on the noun removal sentence, the character string before and after the noun is divided and searched.
The only factor affecting the search time is the number of characters in the target noun or noun removal sentence, and a high-speed search is possible regardless of the number of items included in the database.

  An example of searching for “e1 = he” is shown in FIG. The noun database is searched for the first character “h”, the second character “e”, and the third character “/ (end)”.

  When searching the conditional correspondence database based on the noun removal sentence “E: e1 talk e3 to e2.”, The character string to be searched is the noun removal sentence “blank (e1) talk (e3)”. to (e2). ”, and the search is sequentially performed by dividing into four locations“ / ”,“ talk / ”,“ to / ”,“ ./ ”.

  In machine translation based on the present invention, when a specific noun has a plurality of meanings and judgment by attribute is impossible, a machine translation system displays a plurality of translation results without decision, and allows the user to make a selection decision. Give authority.

  For example, when the English noun “bank” is translated into a Japanese noun, it corresponds to “bank” or “river”, and the attribute of both is “location”. Therefore, when “E: ~ go to the bank ~” is translated into Japanese, for example, “J: go to (bank / river) ~” is displayed, and the user selects one of them. It will be.

Machine translation system configuration diagram Machine translation flowchart Noun database search example

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Input device 200 Output device 300 Translation engine 400 Conditional correspondence type | formula database 410 ・ Conditional correspondence type | formula correspondence database 420 ・ User-specific conditional correspondence type | formula database 500 Noun database 510 ・ Field-specific noun database 520 ・ User-specific noun database

Claims (9)

A machine translation method for translating a source language into a target language,
(1) creating a conditional correspondence expression database for storing a conditional correspondence expression that associates a noun removal sentence in a source language with a noun removal sentence in a target language;
(2) creating a noun database that stores source language nouns, target language nouns corresponding to the source language nouns, and nouns corresponding to the source language nouns and nouns corresponding to the target language nouns; ,
(3) inputting a source language sentence as a translation target;
(4) extracting a noun from the source language sentence and dividing it into a source language noun removal sentence and a source language noun;
(5) searching a noun database based on the source language noun to determine attributes of the source language noun, and translating the source language noun into a target language noun;
(6) Based on the noun removal sentence in the source language, search the conditional correspondence database using the noun attribute as a condition, determine the conditional correspondence expression, and use the noun removal sentence in the source language as the target language Translating into noun-removed sentences,
(7) introducing the target language noun obtained in step (5) into the target language noun removal sentence, and creating a target language sentence;
(8) a step of outputting the target language sentence as a translation result;
A machine translation method comprising:   The conditional correspondence database is a collection of a plurality of conditional correspondence databases classified by style, and the user selects and uses one or more conditional correspondence databases according to the purpose of use. The translation method according to claim 1, wherein the translation method is characterized.   The machine translation method according to claim 1, wherein a user can add a conditional correspondence expression to the conditional correspondence expression database.   The noun database is an aggregate of a plurality of noun databases classified according to a field, and a user selects and uses one or more noun databases according to the purpose of use. The machine translation method described in any one of.   The machine translation method according to claim 1, wherein a user can add a noun to the noun database.   Each item in the conditional correspondence database and the noun database is registered in phonological order, and when searching the conditional correspondence database and the noun database, phonemes are determined for each character in order from the front of the character string. The machine translation method according to claim 1, wherein the machine translation method is performed. A machine translation system that translates a source language into a target language,
(1) an input device for inputting a source language sentence to be translated;
(2) an output device that outputs a target language sentence as a translation result;
(3) a conditional correspondence expression database for storing a conditional correspondence expression that associates a noun removal sentence in the source language with a noun removal sentence in the target language;
(4) a noun database storing source language nouns, target language nouns corresponding to the source language nouns, attributes of the source language nouns and nouns corresponding to the target language nouns;
(5) A translation engine that searches the conditional correspondence database and the noun database based on the input source language sentence and performs translation processing into the target language.
A machine translation system consisting of   The machine translation according to claim 7, wherein the conditional correspondence database, the noun database, and the translation engine are stored in a server, and the server is accessed and used from each communication terminal. system. The machine translation system according to claim 8, wherein the communication terminal is a mobile phone.